Datasets:

CShorten

/

CORD-19-prototype

like 0

Communicable Diseases Prioritized for Surveillance and Epidemiological Research: Results of a Standardized Prioritization Procedure in Germany, 2011

INTRODUCTION: To establish strategic priorities for the German national public health institute (RKI) and guide the institute's mid-term strategic decisions, we prioritized infectious pathogens in accordance with their importance for national surveillance and epidemiological research. METHODS: We used the Delphi process with internal (RKI) and external experts and a metric-consensus approach to score pathogens according to ten three-tiered criteria. Additional experts were invited to weight each criterion, leading to the calculation of a median weight by which each score was multiplied. We ranked the pathogens according to the total weighted score and divided them into four priority groups. RESULTS: 127 pathogens were scored. Eighty-six experts participated in the weighting; “Case fatality rate” was rated as the most important criterion. Twenty-six pathogens were ranked in the highest priority group; among those were pathogens with internationally recognised importance (e.g., Human Immunodeficiency Virus, Mycobacterium tuberculosis, Influenza virus, Hepatitis C virus, Neisseria meningitides), pathogens frequently causing large outbreaks (e.g., Campylobacter spp.), and nosocomial pathogens associated with antimicrobial resistance. Other pathogens in the highest priority group included Helicobacter pylori, Respiratory Syncytial Virus, Varicella zoster virus and Hantavirus. DISCUSSION: While several pathogens from the highest priority group already have a high profile in national and international health policy documents, high scores for other pathogens (e.g., Helicobacter pylori, Respiratory syncytial virus or Hantavirus) indicate a possible under-recognised importance within the current German public health framework. A process to strengthen respective surveillance systems and research has been started. The prioritization methodology has worked well; its modular structure makes it potentially useful for other settings.

The large number of infectious agents with different pathogenspecific, host-specific and socio-economic characteristics makes the allocation of the limited resources available within the area of prevention and control of communicable diseases both challenging and controversial. The amount of attention, efforts and funds spent on surveillance, control and research of infectious pathogens varies greatly between pathogens, settings and over time. This distribution often appears to be ambiguous, potentially guided by senior leaders' research interests, short-term political agenda or residuals of historic situations [1] . As many pathogens are potentially harmful for humans and may present serious public health threats, it is necessary to prioritize the resources dedicated for surveillance and epidemiological research of infectious diseases. This needs to be done sensibly and rationally bearing in mind different aspects of pathogens' characteristics, their impact on societies and long-term consequences of their presence or introduction into populations. The rational and transparent setting of priorities for investment into health research is therefore becoming an essential part of research planning [2] [3] [4] . Usefulness of prioritization, irrespective of its methodology, has been demonstrated by several research groups [5] [6] [7] [8] [9] [10] [11] . Prioritization can provide directions for future resource allocation and strategic planning at different levels (institutional, regional, national or international) and act as a platform for inter-disciplinary debate involving decision-makers, researchers, clinicians and the general public [8, 12] . Although today there are a number of published tools to guide the process of setting priorities, only a few publications openly describe the methodology in sufficient detail and transparency to allow reproducibility or adaptation in other settings [2, 3, 8, 13] . Furthermore, very little is published in terms of actual prioritization results. The Department for Infectious Disease Epidemiology of the Robert Koch Institute (RKI), German national public health institute, is in charge of national surveillance, prevention, control strategies and epidemiological research in the field of infectious diseases. Together with external senior experts the Department initiated a prioritization process aiming to (1) develop a rational system for setting priorities in the area of infectious diseases using a metric-consensus approach, and (2) rank most common pathogens in accordance with their importance for national surveillance and epidemiological research to guide future work of the RKI. In the absence of established standards we designed a methodology using elements of our previous work in 2004 [14, 15] and experiences of other groups [2, 3, 10, 11, 13] . Our multi-staged prioritization process included compilation of the list of pathogens to be prioritized, development of evaluation criteria, weighting of criteria and scoring of the pathogens. The methodology was based on the core domain for good practices in setting priorities for research in health, such as legitimacy and fairness [3, 8] , and represents the further development of the work from 2004. The core team (YB, AG, GK) contacted relevant leading national public health institutions with the request to name a representative to take part in the Delphi consensus process that aimed at assessing the feasibility of the methodology and the relevance of the suggested criteria, as well as to discuss possible improvements and pathogens' scores. Ten external senior experts (BG, UG, JH, TJ, MK, MKr, TL, MP, NS, UU) were nominated. They represented the National Committee of Infectious Disease Epidemiology, the German Society of Hygiene and Microbiology, the German Society of Infectious Disease Specialists, the German Society of Epidemiologists, the National Reference Laboratories, the German Federal Ministry of Health, the RKI Scientific Committee, the German Federal Chamber of Physicians and the German Medical Association. Ten internal experts (GK, AG, YB, SB, RB, TE, OH, KS, OW, MM) represented units and departments of the RKI working in the field of infectious diseases. The participants shared expertise in bacteriology, virology, mycology, parasitology, general infectious diseases, tropical medicine, general medicine, epidemiology, public health, veterinary health and infection control. To maintain a broader approach, we decided to evaluate pathogens rather than diseases for their importance. A list of pathogens was compiled according to the following selection criteria: (a) notifiable according to the German law for the control of infectious diseases [16] , (b) reportable within the European Union [17] , (c) reportable to the WHO under the International Health Regulations [18] , (d) agents with potential for deliberate release [19] , and (e) pathogens represented in dedicated chapters in an established infectious diseases manual [20] and occurring in Germany. Some pathogens were grouped together when biologically and clinically plausible. The list of pathogens was reviewed by the RKI experts and the Delphi process participants. The twelve criteria used during our previous prioritization process in 2004 [14, 15] were further modified according to the feedback received from a broad group of different experts [21] . The newly suggested criteria and their three-tiered definitions were then reviewed by internal and external experts. The initial scoring of all pathogens according to each criterion was performed by the core team and internal experts from the RKI. The data supporting the scoring decisions and references to the data sources, or experts' own explanations were recorded in a structured format. The internal scoring was followed by a modified two-round Delphi consensus process (internal round with RKI and joint round with additional external experts) where scores were discussed. Independent of the scoring, we invited a panel of external experts to assign a weight to each criterion. This invitation was sent to all 16 federal public health institutions, all 18 national reference centers, 49 consulting laboratories, 9 scientific and professional societies working in the area of control, prevention and research of infectious diseases and to all 72 participants of the 2006 online survey [21] . External experts were asked to assign a value ranging from 0 to 10 to each criterion, thus reflecting the criterion's contextual importance for surveillance and epidemiological research. The value 0 reflects the lowest and 10 the highest level of importance of a criterion. More than one criterion could be assigned the same weight, similarly to techniques used in other prioritization exercises [5] . The final criterion's weight was defined by the median value of all weights assigned by the participating experts. Each score was multiplied by the weight for the respective criterion. The sum of these weighted scores reflects the total weighted score of a pathogen. The total weighted scores were finally re-scaled to a range from 0 to 100 in order to facilitate final interpretation. Following the experience from Canada [11] , we did not focus on the exact numerical score assigned to a pathogen. The interpretation of the final weighted scores and their corresponding sequential ranks was done in priority groups reflecting four priority levels (the highest, high, medium and low priority). The cut-off limits for the groups were based on the equal ranges of 25 (0-25, 26-50, 51-75, 76-100). Distribution of the pathogens was later compared with their positions in the 2004 priority list. In total 127 pathogens were selected for prioritization. Drugresistant strains were scored under the common pathogen group and were not assessed as a separate entity. During the Delphi rounds, changes in pathogens' scores were made based on the consensus approach. The Delphi process also resulted in the recommendation to remove the criterion ''Emerging potential'' due to its ambiguous meaning (i.e. each infectious pathogen has an ability to emerge or re-emerge) and to rephrase the definitions of some criteria. The final criteria and their definitions are presented in Table 1 . The detailed scores are available in the Table S1 . All criteria were weighted by a total of 86 experts (14 RKI and 72 external experts). All Delphi discussion participants took part in the weighting. The opinions regarding importance of individual criteria varied greatly across the participants for some criteria, while it was similar for others. For example, eleven participants considered the criterion ''Public attention'' to be of a low importance (scoring it ''2'') and the same number considered it to be of a relatively high importance (scoring it ''6''). At the same time, other criteria such as ''Case fatality rate (CFR)'' and ''Trend'' were weighted in a similar way by the majority of the respondents. When looking at the median weights, the criterion ''Case fatality rate'' was considered the most important criterion (median weight of ''9'') while ''Trend'' and ''Public attention'' (both with a median weight of ''5'') were considered the least important ones ( Table 2) . We analyzed weights assigned by the experts working in different areas of medicine: epidemiologists and public health specialists (n = 43), laboratory specialists (n = 35) and clinicians (n = 8) ( Table 2) . Several criteria were assessed similarly across the groups (for example, the criteria ''Prevention'', median weight of ''8'', or ''Trend'', median weight of ''5''). At the same time, the criterion ''Incidence'' was seen as one of the most important by Need for medical treatment is established but currently no effective treatment is available or AMR limits treatment options AMR = antimicrobial resistance. Note 1. All criteria apply to the geographical settings where the prioritization is conducted; the time-frame applicable to the requested epidemiological data should be defined prior to the process initiation and depend on a frequency with which pathogens are planned to be re-scored. The RKI conducted re-scoring relevant for Germany using a time-frame of 5 years. Indicated numerical thresholds apply to a country where the prioritization process is conducted; when the prioritization is conducted in other geographical settings, different thresholds may need to be considered. Note 2. Event is defined as the occurrence of a disease that is unusual with respect to a particular time, place or circumstances. For certain infectious diseases one case may be sufficient to constitute an event (e.g. polio virus). Public health actions are any kind of targeted actions aiming to identify the nature of the event and/or to apply control measures in response to the event occurrence. *assessed against the total burden of infectious diseases. **assessed for each particular pathogen in question, e.g., for the criterion ''Treatment possibilities and needs'' it therefore refers to availability and adequacy of treatment for each case of an illness caused by a particular pathogen and does not take into account the incidence of illnesses or the availability of preventive measures. doi:10.1371/journal.pone.0025691.t001 epidemiologists and public health experts (median weight of ''8'') and laboratory specialists (median weight of ''7'') but was seen as one of the least important by clinicians (median weight of ''5.5''). Almost the reverse situation was observed with the criteria ''Work and school absenteeism'' and ''Health care utilization'' that were seen to be important by clinicians but perceived as of lower importance by the two other groups. Table 3 presents the allocation of the pathogens into four priority groups according to their weighted total score. The highest priority group contains 26 (20.5%) pathogens. Among those are the pathogens that already received the highest priority in our previous prioritization process, such as HIV, Mycobacterium tuberculosis, Staphylococcus aureus including methicillin resistant strains (MRSA), Influenza virus, Hepatitis C virus, Campylobacter spp., Neisseria meningitidis, Legionella pneumophila, Varicella zoster virus (VZV). It additionally contains a number of pathogens responsible for nosocomial infections (e.g., Klebsiella spp., Pseudomonas ssp., Enterococcus spp.) and Respiratory syncytial virus (RSV) that were not evaluated by us previously. Helicobacter pylori and Hantavirus now belong to this group, while in 2004 both pathogens held medium priority positions. Other pathogens that were scored highest in 2004 were now assessed to be in the medium priority group (e.g., Parvovirus B19). The high priority group contains 39 (30.7%) and the medium priority group -45 (35.4%) pathogens. A number of pathogens that were ranked as the least important in 2004 (Vibrio cholera, Francisella tularensis, Bacillus anthracis, Bartonella quintana) were now assigned to the medium and high priority groups. Out of 17 pathogens from the low priority group in 2011, 11 were newly added. This prioritization approach allowed us to benefit from the cumulative knowledge of many experts. The results of our work seem convincing to us; they support our current activities as well as indicate new directions for the future work. For example, the ranking of the majority of the pathogens found in the highest priority group (e.g., HIV, M.tuberculosis, Influenza virus, Neisseria meningitides, Legionella pneumophila) is in line with strategic goals identified by a number of international agencies that focus both on resource-constrained health systems and on industrialized countries such as Germany [4, 9, [22] [23] [24] . The decision to evaluate a broad range of nosocomial pathogens in the current prioritization and their high ranks indicate a growing recognition of the problem of antimicrobial resistance and healthcare associated infections (HAI) and are in line with a number of new strategic international and national policies calling for enhancement of HAI surveillance activities and capacities [4, 25] . The positioning of Helicobacter pylori, Hantaviruses, RSV and VZV among the pathogens in the highest priority group helped us to identify the under-recognized importance of the diseases with respect to surveillance and epidemiological research and call for actions in this respect. Indeed, although there is a large amount of clinical and laboratory research dedicated to Helicobacter pylori and a number of clinical guidelines are available, very little is done in terms of public health surveillance, despite the growing rates of antimicrobial resistance seriously compromising treatment [26, 27] . RSV remains the most common respiratory pathogen in infants and young children worldwide, often resulting in serious lower respiratory tract infections [28] , yet a routine surveillance system for RSV based on virological testing of sentinel respiratory samples has only been initiated recently, and the burden of disease at the population level in Germany is largely unknown. The placement of RSV in the highest priority group is particularly unusual as pathogens causing diseases in limited population groups, i.e. here in young children, are often believed to be of a lower overall importance. The incidence of human disease caused by Hantavirus fluctuates significantly over time; its incidence has peaked in endemic areas in Germany in recent years, which may have contributed to the higher ranking of this pathogen in the 2011 prioritization and the call to enhance research activities. A population-based seroprevalence survey will indeed be initiated within the RKI-funded network of reference laboratories. VZV, a virus that causes two frequent diseases in children and adults, was ranked as a pathogen with the highest importance both Table 2 . Median weight of each criteria defined by experts from different professional groups (criteria are ranked according to their priority positions among all participants). All participants (n = 86) Area of professional activity [29] . The low priority group contains both pathogens with very low incidence in Germany (e.g., Mycobacterium leprae or helminths) and those much more common (e.g., Roseolovirus or Chlamydia pneumoniae), supporting our approach that the importance of a pathogen is defined by multiple factors rather than by its incidence or prevalence alone. Complex surveillance systems exist for some pathogens that were assigned to the medium priority group, for example, Francisella tularensis or Yersinia pestis. Although the amount of research that should be dedicated to those medium-and lowpriority pathogens must be revised, undoubtedly the need to investigate outbreaks associated with these pathogens, to maintain diagnostic capacity and to continue efficient surveillance to pick up increasing trends of these rare illnesses, remains. Similarly to other research groups that initiated processes for setting priorities [2, 8, 13] , we found the compilation of the data necessary for the scoring process to be challenging due to a limited availability of reliable and published information. Participants did not find it easy to maintain sufficiently broad judgements despite the highly specialized expertise and focused research interest of some experts. Some degree of subjectivity can therefore never be avoided. Another limitation of our study was the difficulty to appropriately account for the decreasing trends of vaccinepreventable diseases that followed successful implementation of effective prevention programmes. Therefore, relatively low scoring of the respective pathogens for some of the criteria should not question the need to maintain adequate surveillance capacity for these illnesses. In the applied methodology, we aimed at following the main principles of good practice in setting priorities in health research [3] and to reach maximum levels of objectivity, transparency, and reproducibility by integrating the following components: 1) a broad but systematic and reproducible selection of pathogens, rather than respective diseases to be included in the prioritization, 2) an explicit definition of each of the ten criteria using a threetiered grading approach, 3) a comprehensive individual pathogenspecific scoring according to the best available evidence reviewed by a multidisciplinary expert group, and 4) a separate weighting of the criteria based on the involvement of a broad range of internal and external experts. Although this approach required intensive preparation, we believe it assured a high level of objectivity and transparency as demanded by Nuyens et al [7] . Our pathogen-specific approach allowed us to conduct prioritization not influenced by programmatic views and to compare pathogens' ranking within a group of diseases as well as across the groups (e.g., a pathogen being targeted by an antimicrobial resistance programme as well as by a zoonoses programme). This approach helped us uncover some differences in the importance of pathogens belonging to the same group, e.g., Neisseria gonorrhoeae and Trichomonias vaginalis. Marked differences were observed in the weighting of the criteria between different professional groups. As the invitation to participate in the weighting was sent to various institutions and often forwarded further, it is not feasible to estimate the response rate. However, we received responses from at least one member of each contacted institution. The weighting of criteria is likely to correlate with societal values and reflect socio-economic, cultural and health system structure in a country. The fact that in Germany the criterion ''Case fatality rate'' was considered to be of the most importance, may reflect the high level of individualization and relative affluence of the German health care system. Comparisons to other countries are difficult since similar studies are lacking. One of the few other priority-setting initiatives with an explicit weighting procedure was conducted in Spain in which the criterion ''Burden and importance of illness'' was identified to be among the first three most relevant from a total of nine criteria, the other two being ''Potential to change health outcomes'' and ''Potential to translate new knowledge into clinical or health services practices'' [5] . We can conclude that weighting needs to be explicitly addressed in any prioritization exercise and it is particularly appropriate for a national public health institute in order to reflect expectations from a broad range of professionals as the respective stakeholders [30] . One way to even expand the societal perspective of the prioritization exercise could be to involve patients' representatives, similarly to how it has been done by Gooberman-Hill et al [31] , for example. Conclusions: The prioritization methodology presented here is based on the systematic evaluation of evidence and the involvement of a broad range of external experts. We feel that the results provide internal consistency and are plausible in the public health perspective. Our comprehensive and transparent approach makes the results defensible and shall give guidance for current needs in surveillance and epidemiological research in Germany. The list of ranked pathogens established here will serve as a reference for our mid-term strategic decisions, which will include strengthening the existing or introduction of new surveillance systems for pathogens from the high priority group (e.g., RSV, VZV or Helicobacterpylori) and re-consideration of the research and surveillance needs for those from the lowest priority group. It has already influenced the decision process on the need for the installation of new and continuation of existing national reference centers in Germany and the internal planning for the respective allocation of resources (GK personal communication). We plan to conduct a re-assessment of priorities within a five-year time frame based on the same methodology. The prioritization tool or its components can be applied across different areas of infectious diseases (by re-weighting prioritization criteria by different professional groups for different purposes) and in different geographical areas (by re-scoring pathogens according to their characteristics relevant for particular countries or continents). We hope that the presentation of our methodology could be helpful to other institutions that choose to prioritize their resources based on a transparent and standardized process.

A Vesicular Stomatitis Virus Replicon-Based Bioassay for the Rapid and Sensitive Determination of Multi-Species Type I Interferon

Type I interferons (IFN) comprise a family of cytokines that signal through a common cellular receptor to induce a plethora of genes with antiviral and other activities. Recombinant IFNs are used for the treatment of hepatitis C virus infection, multiple sclerosis, and certain malignancies. The capability of type I IFN to suppress virus replication and resultant cytopathic effects is frequently used to measure their bioactivity. However, these assays are time-consuming and require appropriate biosafety containment. In this study, an improved IFN assay is presented which is based on a recombinant vesicular stomatitis virus (VSV) replicon encoding two reporter proteins, firefly luciferase and green fluorescent protein. The vector lacks the essential envelope glycoprotein (G) gene of VSV and is propagated on a G protein-expressing transgenic cell line. Several mammalian and avian cells turned out to be susceptible to infection with the complemented replicon particles. Infected cells readily expressed the reporter proteins at high levels five hours post infection. When human fibroblasts were treated with serial dilutions of human IFN-β prior to infection, reporter expression was accordingly suppressed. This method was more sensitive and faster than a classical IFN bioassay based on VSV cytopathic effects. In addition, the antiviral activity of human IFN-λ (interleukin-29), a type III IFN, was determined on Calu-3 cells. Both IFN-β and IFN-λ were acid-stable, but only IFN-β was resistant to alkaline treatment. The antiviral activities of canine, porcine, and avian type I IFN were analysed with cell lines derived from the corresponding species. This safe bioassay will be useful for the rapid and sensitive quantification of multi-species type I IFN and potentially other antiviral cytokines.

IFN-a and IFN-b are structurally related cytokines of the type I interferon family which mediate an early innate immune response to viral infections. There are 13 distinct IFN-a genes present in the human genome, and a single gene encoding for IFN-b. These genes are transcriptionally activated in cells sensing a virus infection through pattern recognition receptors such as the retinoic acid inducible gene I (RIG-I) helicase. Following secretion, IFN-a and IFN-b act similarly by binding to a common, ubiquitously expressed IFN-a/b receptor resulting in the activation of the JAK/STAT signal transduction pathway and transcription of ''IFN-induced genes'' [1, 2] . Several of these genes encode for proteins with strong antiviral activity, i.e. Mx protein, protein kinase R, and 29-59oligo(A) synthetase [3] . Due to their autocrine action, type I IFN may attenuate virus replication in infected cells. Probably more important is the paracrine action of type I IFN, which induces an antiviral state in previously uninfected cells, thereby blocking virus dissemination in the organism. In addition to this ''classical'' antiviral function, type I IFNs are known to affect cell proliferation and differentiation, to modulate the immune response, to inhibit angiogenesis, and to promote apoptosis [4, 5] . Genetically engineered type I IFNs are currently in clinical use for the treatment of multiple sclerosis [6] , chronic hepatitis C virus infection [7] , and certain types of cancer [8, 9] . An issue of increasing importance is the determination of neutralizing antibodies that are induced in some patients following recombinant IFN therapy [10] . Apart from IFN-a/b, cytokines such as IFN-c (type II IFN) and IFN-l (type III IFN) exhibit antiviral activities, although they bind to distinct receptors. In particular, type III IFNs induce transcriptional activation of antiviral genes similar to those activated by type I IFN. Type III IFNs act primarily on epithelial cells [11] and probably play an important role in the innate immune response of epithelial tissues to virus infections [12, 13] . The accurate determination of antiviral IFN activity is a cumbersome issue. In the ''classical'' bioassay, serial dilutions of both a test sample with unknown IFN activity and a type I IFN standard are incubated with an appropriate cell line prior to infection with a cytolytic virus such as vesicular stomatitis virus (VSV), encephalomyocarditis virus, or Sendai virus [14, 15] . The reciprocal value of the highest type I IFN dilution mediating protection of 50% of the cells from virus-induced cytopathic effects (CPE) is defined as one unit of type I IFN per volume. This classical IFN bioassay is time-consuming because the CPE normally needs 24 hours or more to develop. A faster readout can be achieved with recombinant viruses expressing reporter proteins such as green fluorescent protein (GFP) or firefly luciferase [16, 17, 18, 19] . However, any work with live virus requires appropriate biosafety containment. For example, a recently published IFN bioassay can only be performed in biosafety level 3 (BSL-3) facilities, because the test makes use of a recombinant Rift Valley Fever virus [19] . Viral replicon-based bioassays that take advantage of disabled propagation-incompetent viruses may provide an attractive alternative to live virusbased bioassays. A human hepatoma cell line harbouring a selectable hepatitis C virus replicon has been successfully employed for the measurement of type I IFN from patients with chronic hepatitis C virus infection [20] . However, as type I IFNs act in a species-dependent manner, this system may not be applicable to animal IFNs. Transgenic cell lines expressing a reporter gene under control of an IFN-responsive promoter may also be used to determine IFN activity under biosafe conditions [21, 22] . However, it is difficult to simply relate transcriptional reporter gene activation to antiviral activity. In this report, a novel type I IFN bioassay is presented, which is based on BSL-1-classified VSV replicon particles. The assay is highly sensitive and quantifiable due to the expression of a firefly luciferase reporter gene. In addition, the assay can be rapidly performed within 6 to 7 hours and may be used to determine the antiviral activity of IFNs from humans as well as other species. Thus, this bioassay may be of general interest for all those who want to determine the antiviral activity of cytokines such as type I IFNs. Cells BHK-21 cells were obtained from the German Cell Culture Collection (DSZM, Braunschweig, Germany) and grown in Earle's minimal essential medium (EMEM) supplemented with 5% fetal bovine serum (FBS). BHK-G43, a transgenic BHK-21 cell clone expressing VSV G protein in a regulated manner, was maintained as described previously [23] . The porcine kidney cell line PK-15 (ATCC, Manassas, VA) was propagated in Dulbecco's modified Eagle medium supplemented with nonessential amino acids, 1 mM Na-pyruvate and 5% horse serum. D-17 canine osteosarcoma cells (ATCC), Calu-3 human lung adenocarcinoma cells (ATCC), and NHDF normal human dermal fibroblasts (Lonza, Cologne, Germany) were maintained in EMEM with 10% FBS. The UMNSAH/DF-1 (DF-1) chicken fibroblast cell line (ATCC) was maintained in Dulbecco's Modified Eagle's Medium and 10% FBS. All cell lines were cultured at 37uC in a humidified atmosphere containing 5% CO 2 , except DF-1 cells which were kept at 39uC. BALB 3T3 fibroblasts (subclone A31) were kindly provided by N. Pringle, University College, London, UK, and maintained in Dulbecco's Modified Eagle's Medium. Human and murine IFN-b and human IFN-l (IL-29) were purchased from PBL InterferonSource (Piscataway, NJ). Recombinant porcine IFN-a1 [24] was kindly provided by Nicolas Ruggli (IVI, Mittelhä usern, Switzerland). Recombinant chicken IFN-a [25] was kindly provided by Peter Stä heli (University of Freiburg, Germany). Canine IFN-b [26] was kindly provided by Philippe Plattet (Department of Clinical Research and Veterinary Public Health, University of Bern). A plasmid-based rescue system [27] was used to generate a Gdeleted VSV driving expression of firefly luciferase and eGFP. The previously described genomic plasmid pVSV*DG(HA) containing 6 distinct transcription units (N-P-M-HA-eGFP-L) [28] was modified by replacing the influenza virus HA gene in the fourth position with eGFP taking advantage of MluI and BstEII endonuclease restriction sites upstream and downstream of the HA ORF, respectively. The firefly luciferase gene was amplified from the pBI-L plasmid (Clontech) by Pfu PCR and inserted into the fifth transcrition unit using XhoI and NheI endonuclease restriction sites. The resulting plasmid was designated pVSV*DG(Luc) and contained 6 genes in the order N-P-M-eGFP-Luc-L (Fig. 1) . For generation of VSV*DG(Luc) replicon particles, BHK-G43 helper cells were grown in 100-mm diameter culture dishes to 90% confluence and infected with recombinant modified vaccinia virus Ankara (5 pfu per cell) expressing T7 RNA polymerase (MVA-T7, a gift of Gerd Sutter, München, Germany). MVA-T7 has been classified by the German Central Committee for Biosafety as a BSL-1 organism (reference number 6790- [10] [11] [12] [13] [14] . Ninety minutes post infection, the medium was replaced with fresh EMEM containing 5% FBS and 10 29 M mifepristone (Sigma-Aldrich, Buchs, Switzerland) to induce VSV G expression [23] . Subsequently, the cells were transfected with 10 mg of pVSV*DG(Luc), 3 mg of pTM1-N, 5 mg of pTM1-P, and 2 mg of pTM1-L [28] using Lipofectamine TM 2000 transfection reagent (Invitrogen, Basel, Switzerland). The cells were trypsinized 24 hours post transfection and seeded into T75 flasks along with an equal number of fresh BHK-G43 cells. The cells were further incubated at 37uC for 24 hours in the presence of 10 29 M mifepristone. The cell culture supernatant was clarified by low-speed centrifugation and passed through a 0.20-mm-pore size filter to deplete vaccinia virus. VSV*DG(Luc) was further propagated on mifepristoneinduced BHK-G43 cells and stored frozen at 270uC. To determine infectious virus titers, confluent BHK-21 grown in 96well microtiter plates were inoculated in duplicate with 40 ml of serial tenfold virus dilutions for 1 h at 37uC. The wells additionally received 60 ml of EMEM and were incubated for 20 h at 37uC. The infectious titers were calculated according to the number of GFP-expressing cells/well and expressed as fluorescence-forming units per milliliter (ffu/ml). MVA-T7 was titrated on DF-1 cells grown in 96-well microtiter plates. The cells were inoculated with tenfold serial virus dilutions for 90 min and overlayed with medium containing 0.9% methylcellulose. Following incubation for 48 h at 39uC, the cells were fixed with 3% paraformaldehyde, permeabilized with 0.25% Triton X-100, and subsequently incubated with the TW2.3 monoclonal antibody directed to the vaccinia E3L protein (kindly provided by Jonathan Yewdell, NIH, Bethesda, USA) and anti-mouse IgG conjugated to horseradish peroxidase (DAKO). Infected cell foci were visualized with the AEC peroxidase substrate and expressed as plaque-forming units per milliliter (pfu/ml). Using this assay, the final VSV replicon particle preparations proved to be free of MVA-T7. Serial twofold or fourfold dilutions of type I IFN were prepared with cell culture medium containing 5% FBS. The IFN dilutions (100 ml) were added in quadruplicates to confluent cells grown in 96-wells (5610 4 cells/well) and incubated for either 1, 2, 4, or 20 hours at 39uC (DF-1 cells) or 37uC (all other cell lines). The cells were infected with VSV*DG(Luc) (m.o.i. of 5) and incubated for 5 hours at 37uC. The medium was aspirated and 30 ml of luciferase lysis buffer (Biotium Inc., Hayward, CA) was added to the cells. The cell lysates were stored at 220uC. Firefly luciferase activity was determined with a Centro LB 960 luminometer (Berthold Technologies). Luminescence was recorded for 1 s following injection of 30 ml of D-luciferin substrate (Biotium) to white 96-well plates containing 6-ml aliquots of cell lysate. The relative antiviral activity was calculated according to the following formula: Antiviral Activity (%) = 100 -[(RLU +IFN -Blank)6100/ (RLU 2IFN -Blank)]. Mock-infected cell lysates served as blanks and relative light units (RLU) detected with these samples were subtracted from the readings taken from VSV*DG(Luc)-infected cell lysates. RLU +IFN represents the RLU values from IFN-treated cells and RLU 2IFN the readings taken from reference cells, which had not received any type I IFN. A conventional type I IFN bioassay was performed by incubating 96-well cell cultures for 20 hours with twofold dilutions of type I IFN as described above. The cells were infected with propagation-competent VSV (m.o.i. of 1 pfu/cell) and incubated until a cytopathic effect (CPE) was evident in mock-treated control cells. The cells were washed twice with PBS and stained for 1 h with 0.1% crystal violet in 10% formalin. The plates were washed with tap water to remove excess crystal violet and dried. The dye was dissolved by adding 100 ml of 70% ethanol to each well. The absorbance of crystal violet at 595 nm was determined with a microplate reader. The IFN titer was calculated as the reciprocal of the last IFN dilution causing 50% inhibition of virus-induced CPE (50 % reduction of absorbance at 595 nm compared to mock-infected cells) and was expressed as IFN units per volume. Alternatively, antiviral activity was calculated according to the following formula: Antiviral activity (%) = (OD595 +IFN -OD595 2IFN )6100/(OD595 Mock -OD595 2IFN ). OD595 +IFN and OD595 2IFN represent the absorbance of IFN-treated and non-treated cells following infection with VSV and staining with crystal violet, respectively. OD595 Mock denotes the absorbance of non-infected cells. Mean values and standard deviation were calculated. Statistical analysis was performed using the paired Student's t-test. P,0.05 was considered significant. Previously, a VSV replicon vector was generated by replacing the glycoprotein G gene of VSV with the hemagglutinin (HA) gene of an H7N1 influenza virus and inserting an extra transcription unit into the HA-L intergenic junction to drive expression of a modified GFP gene [28] . In the present work, this vector was modified by inserting the firefly luciferase gene into the transcription unit at position 5 (thereby replacing the GFP gene) and exchanging the HA gene with the GFP gene (Fig. 1a) . The resulting vector, VSV*DG(Luc), was propagated on BHK-G43 helper cells which express the VSV G protein in an inducible manner [23] . Up to 10 9 virus replicon particles (VRP) were released into the cell culture supernatant 24 h post infection (data not shown). Several mammalian and avian cells were found to be susceptible to infection with the VRPs in line with previous observations on the very broad cell tropism of VSV [29] . Infected cells were readily detected (5 to 6 h post infection) by means of GFP reporter expression (Fig. 1b) . However, non-helper cells were unable to complement the VSV G deletion and thus could not produce infectious progeny virus. Importantly, VSV*DG(Luc) did not induce type I IFN in the cell lines analysed (data not shown). This can be ascribed to the host shut-off activity of the VSV matrix protein [30] . Infection of BHK-21 cells with VSV*DG(Luc) using a multiplicity of infection (m.o.i.) of 4 ffu/cell resulted in increasing firefly luciferase reporter activity with time (Fig. 1c) . In all subsequent experiments, luciferase activity was generally recorded at 5 h post infection as the signal-to-noise ratio was sufficiently high at this time. When BHK-21 cells were infected with different virus dose rates, luciferase reporter activity increased linearly between 0.004 and 40 ffu/cell (Fig. 1d) The luciferase reporter activity in VSV*DG(Luc)-infected cells depends on the replication/transcription levels of the RNA replicon. Thus, a reduction in viral genome replication as a consequence of type IFN action should lead to correspondingly lower luciferase levels. To test this hypothesis, normal human dermal fibroblasts (NHDF) were incubated for various time periods with serial dilutions of human IFN-b before infection with VSV*DG(Luc). A dose-dependent effect of IFN-b on firefly luciferase reporter activity was observed after 1 hour of incubation (Fig. 2a) . About 1 unit of IFN-b led to 50% suppression of luciferase activity. Extending the incubation time to 2 hours did not further improve the sensitivity of the test (p.0.05 for the 0.08 to 1.25 IFN units range). However, if the cells were treated for 20 hours, the dose response curve was significantly shifted to lower IFN units (p#0.001 for the 0.02 to 20 units range). Approximately 0.05 units of IFN-b were now sufficient to suppress reporter activity by 50%. These results indicate that an IFNb-induced antiviral state in NHDF cells can be detected as early as 1 hour after addition of IFN-b, although the sensitivity of the assay is higher if the cells were incubated with IFN-b for prolonged time. VSV*DG(Luc) was also used to quantify the activities of porcine and chicken IFN-a. While the dose response curve of chicken IFNa on DF-1 chicken fibroblasts (Fig. 2b) was similar to the one of human IFN-b on NHDF (Fig. 2a) , porcine IFN-a showed a different kinetics on PK-15 porcine kidney cells. In these cells, a full antiviral state was accomplished only after 20 hours of treatment indicating that PK-15 cells respond rather slowly to the action of type I IFN (Fig. 2c) . Thus, the assay is applicable to type I IFN from different species but may perform differentially on distinct cell types. To further evaluate the bioassay, samples containing type I IFN of unknown activity were tested against a commercial IFN-b standard and the results were compared with those obtained with a conventional IFN bioassay based on VSV cytotoxicity. Human type I IFN was induced in NHDF human fibroblasts following infection with VSV*M q [30] . This propagation-competent virus expresses a mutant matrix protein that is unable to block the nucleocytoplasmic RNA transport of the cell. Before the samples were tested for antiviral activity, VSV*M q was inactivated for 30 min with 0.1 M HCl to avoid any interference with the bioassay (see also the section on thermal and pH stability of IFN). When the VSV*DG(Luc) replicon bioassay was used, about 0.8 pg/ml of a commercial IFN-b standard resulted in 50% antiviral activity. In consideration of the final dilution producing 50% antiviral activity, the type I IFN concentration of two test samples was defined as 4000 pg/ml and 1440 pg/ml, respectively (Fig. 3a) . When the samples were tested with the conventional bioassay (Fig. 3b) , the values were in the same range (5000 and 1500 pg/ml), indicating that both assays principally agree. Nevertheless, the replicon-based bioassay proved to be more sensitive than the conventional one as lower amounts of type I IFN were sufficient to reduce the luciferase reporter activity by 50% (compare curves shown in Fig. 3a with those in Fig. 3b ). The species-dependent action of type I IFN was analysed by incubating human, canine, murine, and chicken cells for 20 hours with 5 units of type I IFN from either the homologous species or a different one. When cells were treated with the respective homologous IFN an antiviral state was induced as indicated by the lack of GFP expression 6 hours post infection with VSV*DG(Luc) (Fig. 4a) . In contrast, untreated cells or cells that had received type I IFN from a different species were not protected and showed GFP expression accordingly. To compare the effects of different concentrations of homologous and heterologous type I IFN on human cells, we treated NHDF for 20 hours with serial dilutions of human, porcine, and murine type I IFN prior to infection with VSV*DG(Luc). Firefly luciferase reporter activity in infected cell lysates indicated that porcine IFNa is active in NHDF albeit at reduced levels compared to human IFN-b (Fig. 4b; p#0 .0049 for the 0.008 to 2.0 units range). Murine IFN-b showed an even lower activity on NHDF (ED 50 of 500 units per 5610 4 NHDF cells; p#0.0003 for the 0.008 to 500 units range). These results demonstrate that type I IFNs act in a speciesdependent manner and accentuate the need for selecting appropriate cell lines to determine their activity. We often encounter the problem that the activity of type I IFN has to be determined in a sample containing live virus. As virus infection may interfere with the bioassay, it has to be inactivated before the assay is performed, preferentially without touching the activity of type I IFN. As many viruses can simply be inactivated by treatment with heat, we first analysed the thermal stability of human IFN-b using the VSV*DG(Luc) bioassay. The antiviral activity of human IFN-b was fully maintained when the cytokine was incubated for 30 min at 50uC (Fig. 5a) . At 60uC, 99% of the activity was preserved (p = 0.021). At 70uC, the activity of human IFN-b dropped to 83% (p = 0.0017). Incubation at higher temperatures (80uC and 90uC) affected the activity more drastically, although some activity was still left at these temperatures. Only when human IFN-b was heated to 100uC for 30 min, activity was completely abolished. In contrast to human IFN-b, porcine IFN-a was completely inactivated at 70uC (p,0.0001), whereas human IFN-l (IL-29), a type III IFN, showed even higher residual activity at 80uC and 90uC (p,0.001), suggesting that these cytokines have different physicochemical properties. To study the sensitivity of type I and type III IFNs to conditions of extreme pH, human IFN-b and human IFN-l were treated for 30 min with either 0.1 M HCl, 0.1 M NaOH or H 2 O, adjusted to neutral pH, and assayed on NHDF and Calu-3 cells, respectively. It turned out that the antiviral activity of human IFN-b was not significantly affected by acid (p.0.05) (Fig. 5b) , confirming the previously noted acid-stability of type I IFNs [31] . IFN-l showed similar properties as antiviral activity was maintained following treatment with 0.1 M HCl (Fig. 5c) . In contrast, alkaline treatment significantly reduced the activity of IFN-l (p,0.05 for the 250 2 1 ng/ml range), whereas IFN-b was not affected (p.0.05) (Fig. 5b) . Thus, treatment with acid may be employed to inactivate virus in a test sample without affecting the activity of type I and type III IFNs. On the other hand, alkaline may be used to differentiate between type I and type III IFNs. Conventional bioassays take advantage of the antiviral activity of type I IFN to measure the inhibition of virus-induced cytopathic effects in cell culture [14] . In this study, we presented an improved bioassay by employing a recombinant VSV replicon equipped with two reporter proteins. This assay proved to be advantageous over the conventional assay with respect to biosafety, sensitivity, and time requirements. The use of cytopathic viruses in conventional type I IFN bioassays makes appropriate biosafety measures necessary to reduce the risk of unwanted virus transmission and infection. In this regard, the VSV*DG(Luc) replicon particles can be regarded as biosafe. Since the modified VSV genome lacks the envelope glycoprotein (G) gene, VSV*DG(Luc) is propagated on helper cells providing the G glycoprotein in trans [23] . The replicon particles produced on these cells can run a single cycle of infection but cannot produce any infectious progeny [28] . Another aspect contributing to biosafety is that the RNA replicon replicates exclusively in the cytosol and does not produce cDNA intermediates. Thus, in the case of an accidental infection any risk of recombination with or integration into host chromosomal DNA can be excluded. Virus preparations that are used for IFN bioassays must not contain any type I IFN which may distort the results. In addition, the viruses should not induce IFN in infected reporter cells to assure that the effects measured are solely due to the exogenous IFN added. Both criteria are fulfilled for VSV*DG(Luc). The helper cells used to propagate the replicon particles are derived from BHK-21 cells, which are defective in the synthesis of type I IFN [32] . In addition, the VSVDG replicon does not induce type I IFN in infected reporter cells, because the VSV matrix protein efficiently blocks the nuclear export of cellular mRNA including the IFN mRNA [30, 33] . Conventional type IFN bioassays often take 24 hours or more to be completed as they rely on the inhibition of cytopathic effects. In contrast, the bioassay presented here takes advantage of the firefly luciferase reporter, which can be detected with high sensitivity long before a cytopathic effect is apparent. Although the luciferase reporter enabled us to unambiguously quantify the IFN-mediated reduction of virus replication, we frequently observed that the standard deviations for the antiviral activities increased when the cells were treated with low amounts of type I IFN. This likely reflects the high reporter gene expression at low levels of type I IFN and the inaccuracy associated with pipetting small volumes (6 ml). The engagement of a common cellular receptor by type I IFNs leads to activation of the JAK/STAT pathway and transcriptional induction of several genes with antiviral activity [1, 2] . It is common practice that cells are incubated with type I IFN for several hours to induce an antiviral state [15, 19] . Our findings suggest that an antiviral state can be detected much earlier. For example, treatment of NHDF with 1 unit of human IFN-b for 1 hour and subsequent infection with VSV*DG(Luc) for 5 hours was sufficient to suppress firefly reporter expression by 50%, even though the effective dose was further lowered with longer incubation times. While DF-1 fibroblasts responded to chicken IFN-a with dose response kinetics similar to the one observed with human IFN-b on NHDF, PK-15 cells responded to porcine type I IFN in a much slower way. Thus, compared to conventional bioassays VSV*DG(Luc) may allow quantification of type I IFN in a considerably shorter time provided an appropriate cell line has been selected. The selection of a suitable cell line is also important with respect to the species it is derived from, as full activity of type I IFN was only observed with cells from the corresponding species. Since VSV is able to infect a broad spectrum of mammalian and avian cells, the new bioassay may be used to determine the bioactivity of type I IFNs from many different animals. Although type I IFNs are the predominant cytokines with antiviral activity [34] , it cannot be excluded that test samples contain other cytokines that inhibit VSV replication. Indeed, using the VSV*DG(Luc) replicon assay, it was possible to quantify the antiviral activity of IFN-l (Il-29), a type III IFN. This raises the important question of how to distinguish between different antiviral cytokines? Certainly, neutralizing antibodies may be employed to specify the antiviral cytokine present in the sample. In addition, the extraordinary acid-stability of type I IFNs may be used to differentiate them from acid-labile cytokines such as type II IFNs [35] . However, acid may not be used to distinguish between type I and type III IFNs as IFN-l proved to be acid-stable as well. Since IFN-l but not IFN-b was sensitive to 0.1 M NaOH, treatment with alkaline may be used instead. However, the conditions of treatment still have to be optimized to guarantee the complete inactivation of type III IFNs while maintaining the activity of type I IFNs. In addition to antiviral cytokines, test samples may also contain unknown viruses that potentially interfere with the bioassay. Treatment with heat or acid may be used to inactivate these viruses without affecting the bioactivity of type I and type III IFNs. However, as heat-inactivated influenza viruses still may be able to induce type I IFN [36] , virus inactivation by acid may be more convenient. The correct and reliable determination of type I IFN biological activity is an important issue. The new bioassay presented in this study represents an attractive alternative to conventional type I IFN bioassays that work with cytotoxic live viruses. VSV*DG(Luc) is easily produced and handled under BSL-1 conditions. The activity of the replicon can be easily determined and standardized taking advantage of the two reporter proteins. It may be stored for prolonged time in lyophilized form without losing its activity. Finally, the proven sensitivity, rapidity, and accurateness of the assay recommends it for the determination of type I IFN from a number of mammalian and avian species. Finally, the test may be further developed to quantify the antiviral activity of other cytokines such as type II and type III IFNs.

Characterization of Neutralizing Profiles in HIV-1 Infected Patients from whom the HJ16, HGN194 and HK20 mAbs were Obtained

Several new human monoclonal antibodies (mAbs) with a neutralizing potential across different subtypes have recently been described. Three mAbs, HJ16, HGN194 and HK20, were obtained from patients within the HIV-1 cohort of the Institute of Tropical Medicine (ITM). Our aim was to generate immunization antibodies equivalent to those seen in plasma. Here, we describe the selection and characterization of patient plasma and their mAbs, using a range of neutralization assays, including several peripheral blood mononuclear cell (PBMC) based assays and replicating primary viruses as well as cell line based assays and pseudoviruses (PV). The principal criterion for selection of patient plasma was the activity in an ‘extended incubation phase’ PBMC assay. Neutralizing Abs, derived from their memory B cells, were then selected by ELISA with envelope proteins as solid phase. MAbs were subsequently tested in a high-throughput HOS-PV assay to assess functional neutralization. The present study indicates that the strong profiles in the patients' plasma were not solely due to antibodies represented by the newly isolated mAbs. Although results from the various assays were divergent, they by and large indicate that neutralizing Abs to other epitopes of the HIV-1 envelope are present in the plasma and synergy between Abs may be important. Thus, the spectrum of the obtained mAbs does not cover the range of cross-reactivity seen in plasma in these carefully selected patients irrespective of which neutralization assay is used. Nevertheless, these mAbs are relevant for immunogen discovery because they bind to the recombinant glycoproteins to which the immune response needs to be targeted in vivo. Our observations illustrate the remaining challenges required for successful immunogen design and development.

Despite intense research efforts over nearly three decades, only minimal progress has been made in developing an HIV-1 vaccine. In retrospect, a number of reasons can be proposed for this failure such as the enormous genetic diversity of HIV, the camouflage of the neutralizing epitopes in the envelope spike by glycan shields, the presence of ''decoy'' immunodominant non-neutralizing antigenic determinants in non-conserved areas on the surface and the low gp120 trimer spike density on the virus membrane [1] . In addition, the most vulnerable regions may only be accessible for a short period. These short-lived structures include the so-called CD4 induced (CD4i) in gp120 and the pre-hairpin epitopes in gp41 that are only exposed following CD4 receptor binding and the subsequent conformational changes. Still, a few antibodies (Abs) are able to successfully interfere with the binding and fusion process, as seen in passive immunization studies in the macaque model. Such mAbs include 2G12 (binds to mannose residues on gp120); b12 and F105 (bind to the CD4 binding site, CD4bs); 17b and 65 (recognize conformational epitopes in the CD4i region); and 4E10 and 2F5 (bind to epitopes in the membrane proximal extracellular region or MPER of gp41). Last year, however, three new mAbs (HJ16, HGN194 and HK20) were reported from African patients from the ITM HIV-1 cohort. Taken together these mAbs target three different steps in viral entry: binding to CD4bs and thus preventing interaction of HIV-1 with CD4 by HJ16, binding to V3 and blocking the coreceptor binding by HGN194 and finally immobilizing the unfolding of the gp41 by the HK20 mAb [2] . Since HK20 targets HR1 instead of MPER or glycans in this region, it has the conceptual advantage over 4E10 and 2F5 of avoiding potential auto reactivity [2, 3] . Importantly, the HGN194 mAb has recently been found to confer protection in infant rhesus monkeys by the group of Ruprecht [4] . In order to generate these mAbs, patient plasma were selected with a neutralization assay with an extended incubation time, using activated PBMC and a panel of clinically isolated replication competent HIV-1 strains. This assay differs from the classical 'short' PBMC neutralization assay by extending the incubation phase of plasma with virus from 1 to 24 hours. The importance of this format was shown in a SHIV challenge trial in rhesus macaques, where recombinant HIV envelope immunizations induced protection [5, 6] . Comparing various neutralization assays, we showed that the PBMC based assay with an extended incubation phase was able to discriminate between protected and non-protected animals after vaccination. Since we are attempting to develop a vaccine effective against a range of subtypes and because the subtype A, subtype C and circulating recombinant form (CRF) 02_AG are responsible for at least 75% of the current new infections worldwide, we identified patients, whose plasma could cross-neutralize mainly viruses from these three subtypes in the extended incubation PBMC assay. From the blood of selected patients, memory B cells were isolated and immortalized using an Epstein Barr Virus (EBV) based procedure [7] . Supernatants of B cell clones were tested in ELISA with recombinant gp41, trimeric gp120 and gp140 proteins from several subtypes as solid phase. Clones with binding activity to any of these antigens were expanded and supernatants were tested using a HOS based PV neutralization assay. This effort ultimately resulted in the selection of the new mAbs HJ16, HGN194 and HK20, which showed considerable breadth of neutralizing activity against a panel of HIV-1 primary isolates spanning both tier 1 and tier 2 viruses of different subtypes [2] . Here, we present the characteristics of the patient's plasma and their respective mAbs in multiple neutralization assay formats. The results clearly demonstrate that patient selection was highly dependent on the neutralization assay. Although the crossneutralizing properties of the isolated Abs showed considerable variation with the neutralization assay format, all assays indicate that neutralizing Abs to other epitopes of the HIV-1 envelope are present in the plasma and also do not exclude the role that synergy between such Abs could play. The study was approved by the Institutional Review Board of the Institute of Tropical Medicine and the Ethical Committee of the University Hospital of Antwerp. All participants understood and signed an informed consent. Eligible patients visiting the ITM clinic in Antwerp had been infected for at least one year, were clinically asymptomatic and over 18 years old. Neither CD4 T cell counts nor viral loads were taken into consideration. Patients were preferentially selected from sub-Saharan regions where the subtypes A, C and/or CRF02_AG are prevalent. Plasma was subsequently screened for its ability to neutralize a panel of four subtype A, four subtype C and six CRF02_AG primary HIV-1 strains, in our extended incubation phase PBMC assay (see below). HJ16, HK20 and HGN194 Abs were obtained as part of the Collaboration for AIDS Vaccine Discovery program from Dr. D. Corti (Institute for Research in Biomedicine, Bellinzona, Switzerland). Buffy coats from healthy donors from the Red Cross Blood Transfusion Center at the University Hospital of Antwerp were used for isolation of PBMC by LymfoPrep (Axis-Shield, Oslo, Norway) centrifugation and adjusted to 1610 6 /ml in culture medium, consisting of RPMI 1640, 15% fetal calf serum (FCS), 0.03% L-glutamine and 50 mg/ml gentamycin (Lonza, Verviers, Belgium), 2 mg/ml polybrene (Sigma-Aldrich, Bornem, Belgium). Cells were stimulated with 0.5 mg/ml phytohemagglutinin (PHA, Oxoid, Hampshire, UK) for 2 days and 1 day with 200 U/ml interleukin-2 (IL-2; Gentaur, Brussels, Belgium) in a 7% CO 2 incubator at 37uC and then used for neutralization assays. The following cell lines were obtained through the NIH AIDS Research and Reference Reagent Program, Division of AIDS, NIAID, NIH: TZMbl from Dr. John C. Kappes, Dr. Xiaoyun Wu and Tranzyme Inc. and Ghost (3) . All isolates were classified by phylogenetic analysis of their envelope genes. All virus stocks were prepared and titrated on PHA/IL-2 stimulated PBMC. These strains have been extensively used for at least 10 years at ITM and are considered equivalent to neutralization resistant tier 2 viruses [8, 9] . Corresponding envelope PV constructs were obtained by DNA amplification of the complete env starting from PBMC co-cultures or by RT-PCR using plasma and subsequent cloning into an expression vector (pSV7d or pcDNA4/TO) [10] . These included the ITM strains VI 191 (A), VI 829 (C), VI 882 (C), VI 1358 (C), VI 824 (D), VI 1888 (CRF01), VI 1090 (CRF02), CI 20 (CRF02) and CA 18 (CRF02). The env expressing plasmids 92RW009 (A), SF162 (B) and 92BR025 (C) were provided by the EU Programme EVA Centre for AIDS Reagents, NIBSC, UK (AVIP Contract Number LSHP-CT-2004-503487). Sequencing of the PV constructs and phylogenetic analyses of the complete gp160 confirmed the identity of the PV and its corresponding virus. Since several parameters influence the observed neutralizing profile of a plasma or mAb, we included a comprehensive range of different neutralization assays with distinctive characteristics. Apart from the difference in target cell (primary cells vs. cell lines), incubation, absorption and culture phases were also investigated as determinants of neutralization outcome. Formats for the different assays are as shown in Table 1 . PBMC based assays. All PBMC neutralization assays are described as a/b/c where 'a' is the incubation time in hours following mixing of mAb with virus, 'b' is the absorption time in hours during which the cells are exposed to the mAb/virus mixture. Cells are then washed and 'c' is the culture time in days (all at 37uC and 7% CO 2 ). In this study results were obtained in 24/1/14, 1/2/7 and 1/24/14 formats. These are named 'extended incubation', 'short incubation' and 'extended absorption' assays respectively. The extended incubation assay, which was originally used for patient selection has been described previously [6, 11] . Briefly, virus stock is diluted in a five fold series from 1/2 to 1/6250 in culture medium (RPMI-1640 medium supplemented with 15% FCS and 200 U/ml IL-2) to establish the titer (-log10 if the dilution at which 50% infection is achieved). A titer below 1 constitutes poor growth of the virus and the experiment is discarded. Ninety ml of each virus dilution are mixed with 5 ml of plasma or 50 mg mAb. In assays testing neutralization by plasma the mixture is complemented with 5 ml culture medium to give a final 1 in 20 dilution of plasma. When testing mAb, the mixture is complemented with 5 ml flow through (IgG was removed from HIV-1 negative plasma using a Protein G column [GE Healthcare Europe GmbH, Belgium]) to give 50 mg/ml of mAb. After the incubation phase 20 ml of each plasma or mAb/ virus mix are first dispensed in quadruplicate into flat bottom 96well microplates and 75,000 PBMC in 100 ml culture medium are added to each well. Plates are then left in a CO 2 incubator at 37uC during the absorption phase (b). Afterwards, cells are washed three times by centrifugation at 2000 rpm for 10 minutes, the supernatant is aspirated and 180 ml fresh culture medium are added to the cells. When cultured for 14 days, 125 ml of the medium is aspirated and replaced with 135 ml fresh culture medium. After c days, 200 ml of the supernatant are mixed with 50 ml Nonidet P40 (0.25% in PBS; Fluka, Sigma-Aldrich, Puurs, Belgium) to disrupt virions and this mixture is analyzed for the presence of HIV p24 antigen. As a control, pooled plasma of 100 HIV-1 negative donors are tested in parallel. For the short incubation and extended absorption phase assays, times of incubation are appropriately adjusted: 1/2/7 and 1/24/14 respectively. Neutralization activities are presented as the percentage reduction in infectious titer of a virus isolate following incubation with patient plasma or mAb relative to its titer following incubation with HIV-1 negative control plasma. Virus titers were calculated by the method of Reed and Muench [12] . An 80% reduction in titer was considered significant. By extending the usual one hour absorption phase of the extended and short incubation PBMC assay to 24 hours (1/24/14 format) we aimed to reproduce the conditions of the cell line based assays where the mAbs remain during the entire absorption and culture phase. Pseudovirus based assays. Neutralization capacity of patient plasma and mAbs against PV on TZMbl and the HOS cell related GHOST.CD4-X4/R5 cells was determined as described [13, 14] . Luciferase reporter gene activity was quantified 48-72 h after infection upon cell lysis and addition of firefly luciferase substrate (Perkin-Elmer) as described. Emitted relative light units (RLUs) were quantified on a LB941 Berthold luminometer (Alabama, US). Infection of TZMbl cells was quantified using SteadyLite and infection of GHOST cells was quantified using BriteLite as a substrate (both Perkin-Elmer). In a preliminary experiment 1.10 4 TZMbl or GHOST cells were seeded in each well of 96-well, flat bottom plates and infected with a range of viral doses in a total volume of 200 ml to establish the dose, which resulted in a signal of 50,000 to 100,000 RLU in the presence of 10 mg/ml diethylaminoethyl-dextran (DEAE-dextran, Sigma, Belgium) to enhance virus infectivity in TZMbl cells, while no DEAE-Dextran was used in GHOST cells. In the actual neutralization experiments, mAbs or plasma were pre-incubated with PV for 1 h at 37uC. The mAb concentration or plasma dilution producing a 50% reduction in luciferase reporter gene production was determined by linear regression analysis in Microsoft Office Excel as described on http://www.hiv.lanl.gov/content/nab-reference-strains/ html/Protocol-for-Neutralizing-Antibody-Screening-Assay-for-HIV-1in-TZM-bl-Cells-November-2010.pdf. For IC50 of mAbs, the 50% inhibitory concentrations were determined via a linear interpolation method using the mean of duplicate or triplicate cultures. The assay readouts for the dilutions above and below the IC50 were joined with a straight line, plotted against the log concentration of mAb. The position where the line crossed the 50% assay readout was taken as the IC50 estimate. Where the IC value was outside the range of concentrations tested, it was recorded as either greater than the highest concentration used, or less than the lowest concentration, as appropriate. An ID50 for plasma and IC50 for Abs were calculated from a dilution series starting from 1:20 for plasma and starting from 50 or 150 mg for Abs depending on the Ab used. The virus titer was calculated within each individual experiment using the method of Reed and Muench [12] . In the virus dilution series, doses ranged between those infecting all cultures (100%) to those infecting none (0%). Wells giving an OD.0.3, against a background of 0.03-0.05 in the ELISA, were considered to be infected. The infectious virus titer was calculated following virus incubation with mAb/plasma. The reduction in titer was calculated as a percentage of the virus titer following exposure to either IgG or plasma which was pooled from 100 HIV-1 negative donors. Purified IgG from this pool was used as the control for mAbs. Correlations were calculated using the Spearman Rank correlation test using Prism version 5.0. Differences or correlations between sets of data were considered significant if p#0.05 and r.0.5. Over 1400 HIV-1 infected individuals regularly attend the clinic at ITM. Of these, 200 patients were identified whose origin was the sub-Saharan regions of Africa where subtype A, subtype C and/or CRF02_AG isolates are prevalent. Their plasma was evaluated when they were therapy naïve or at least 6 months therapy-free for the ability to neutralize primary HIV-1 strains from the A, C and/or CRF02_AG subtypes in the 24/1/14 extended incubation phase PBMC neutralization assay. About 25% of these patients had cross-neutralizing plasma i.e. plasma which neutralized at least 50% of strains belonging to one subtype plus at least 25% of strains from a second. We next classified the best responding plasma according to the HIV subtype they preferentially neutralized: e.g. when at least three out of four of the A or C strains (or five out of six CRF02 strains) gave greater than 80% neutralization. In Table 2 this neutralization profile is shown for 20 patients whose memory B cells were interrogated. According to these criteria, four patients' plasma preferentially neutralized subtype A strains (HGL-, HGD-, HQ-and HGN plasma) and two were more specific for subtype C (HVDA and HK plasma). We did not find any patients preferentially recognizing the CRF02 strains. Three of the tested patients neutralized subtypes A and C more than CRF02 (HMB-, HJ-and HGR plasma), one was more C and CRF02 subtype specific (HMQ plasma) and finally five patients displayed broad crossneutralizing activity over all three subtypes (HU-, HP/HM/ HGM-, HE-, HY-and HMV plasma). The remaining five interrogated patients did not display this subtype specific behavior (HL-, HZ-, HGP-, HR-and HMA plasma). There is no obvious association between the subtype infecting a patient and that neutralized by his or her plasma. The patients, from whom the newly isolated mAb were obtained, are underlined in the first column. Remarkably, plasma from these patients showed a rather subtype specific neutralization profile since plasma from patient 242315 (HJ patient) neutralized mainly A and C strains, plasma from patient 314994 (HGN patient) mainly A strains and plasma from patient 529552 (HK patient) mainly C strains. In Table 3 the clinical histories of these patients are summarized. Patient 242315 from whom the CD4bs specific HJ16 mAb was obtained was a 45 year old Congolese woman who had been visiting our clinic since 1996. She received treatment intermittently and consequently had a varying CD4 count and viral load. Her neutralization profile had been obtained using plasma samples taken after stopping anti-retroviral therapy for 6 to 11 months but she was back on therapy at time of memory B cell interrogation for 7 months. Patients 314994-HGN and 529552-HK were not receiving antiretroviral treatment during this study. Patient 314994 from whom the V3 crown specific mAb HGN194 was obtained was a 41 year old woman from the Republic of Guinea who has been regularly attending our clinic since 1998. She has always maintained low viral loads and high CD4 T counts so far without treatment. Her viral loads varied between undetectable and 2,700 RNA copies/ml while her CD4 counts have fluctuated between 550 and 960 cells/ml. Patient 529552 whose HK20 mAb is specific for the HR1 region of gp41 was a 31 year old Ghanaian woman. Soon after arrival in Belgium in 2005 she tested positive for HIV. Her viral loads (1.500-40.000 RNA copies/ml) have Table 2 ). In view of the known neutralization resistance of these isolates [6, 8] they were considered to represent 'Tier 2 like'' strains. In the experiments represented in Table 4 , an additional panel of four subtype B, four subtype D and four CRF01_AE strains provided us with an overview of the neutralizing potential of the three selected patient plasma across six subtypes with a total panel of 26 ''tier 2 like'' strains. As can be observed, the 242315-HJ patient plasma has a very broad neutralization spectrum with 21 of the 26 viruses neutralized, including all the C and CRF01 strains, 75% of the subtype A, B and D strains and 67% of CRF02 strains. The 314994-HGN plasma has a narrower range, neutralizing 13/26 viruses, including all of the B strains, 75% of the A strains, 67% of the CRF02 strains, 50% of C strains, but none of the D nor CRF01 strains. The 529552-HK plasma neutralized 12/26 viruses, including 75% of the C strains, 67% of the CRF02 strains, 50% of the A strains and 25% of the B, D and CRF01 strains. Influence of neutralization assays on plasma neutralization profile of the three patients from whom the new antibodies were isolated In order to illustrate the influence of different assays on the neutralization spectra of these selected plasma, we compared results from all the assays shown in table 1 (except for the HOS-PV assay). The virus panel used in this comparison consisted of nine strains from our primary selection panel from which PV were also available. Three strains from the standardized ''NeutNet'' panel were added: A (92RW009, tier 2), B (SF162, tier 1A) and C (92Br025, tier 2) [15] and personal communication). Results are shown in Table 5 . Comparing the neutralization breadth of the three patient plasma in three variants of the PBMC assay, indicates that the HJ and HK plasma neutralize much fewer viruses and the HGN plasma even loses all significant neutralization capacity in the classical short assay (1/2/7), implying that none of them would have been selected using results from this assay. Prolonging the absorption phase to 24 hours (1/24/14), to more closely resemble the cell line based assays (see table 1), only ''rescues'' some neutralization with the HGN plasma. No correlation was found between the results obtained in the different PBMC assays using the Spearman Rank correlation test. There was a correlation between results from the 24/1/14 extended incubation PBMC assay and those with the TZMbl assay using replication competent ''primary'' viruses for the 242315-HJ plasma (r = 0.62, p = 0.03) but not for the other 2 plasma samples. A stronger correlation (r.0.60 for all three plasma) was found for the 242315 and 314994 plasma between the 24/1/14 PBMC assay and the TZMbl_PV assay. The correlation was statistically significant (p,0.04). The strongest correlation (r.0.69) was observed between the two PV assays (TZMbl-PV and GHOST_PV). Correlations for all three plasma were significant (p,0.01). Evaluation of plasma vs. antibodies in the 24/1/14 extended incubation PBMC assay. The neutralization profiles of the plasma are compared with those for their respective mAbs for the 24/1/14 extended incubation PBMC assay in Table 6 . The mAbs clearly neutralized a much more restricted range of isolates than the plasma. 242315-HJ plasma and HJ16 mAb both neutralized the subtype C isolate VI829, the subtype D CI 13 and two of the three CRF02_AG isolates, VI 1090 and CA18. 314994-HGN plasma and HGN194 mAb as well as 529552-HK plasma and HK20 mAb neutralized SF162 (B) and 92Br025 (C) while HGN194 mAb also neutralized VI 191 (A) and 89.6 (B). In Table 5 . Neutralization profile of patient plasma in different HIV-1 neutralization assays. addition, a number of qualitative discrepancies were observed in that growth of some viruses was strongly inhibited by the mAb, but enhanced by the plasma (e.g. subtype A 92RW009 with 242315-HJ plasma versus HJ16 mAb) or vice-versa (e.g. CRF02 strains with 314994-HGN plasma versus HGN194 mAb). This type of inconsistency was also observed by the group of Nussenzweig when they compared their new mAbs with results from the original plasma [16] . Some neutralization-sensitive isolates, CA1 (A), MN (B), BaL (B) and CI 13 (D), were added to the panel in table 6. However, there was only a limited increase in range with HJ16 mAb reaching 80% neutralization against CI 13 (D) and HGN194 mAb almost neutralizing BaL (B) to 80%. HK20 mAb had no activity against any of the extra isolates. While the plasma demonstrated their broadest range of neutralization in the 24/1/14 assays they also showed activity in the other PBMC and cell-line assays. Since it is possible that the mAbs could share these activities we extended their evaluation to assays with the different formats ( Table 7) . The range of HIV-1 isolates neutralized by both the plasma and mAbs is greatest in the extended incubation 24/1/14 PBMC assay while only three of the 36 mAb/isolate combinations show significant neutralizing activity in the extended absorption 1/24/14 PBMC assay. The HJ16 mAb neutralizes three (SF162, VI 1888 and VI 1090) of the six isolates (92RW009, 93Br025 and CA18) neutralized by the plasma in 1/2/7 PBMC assays (table 5). The HGN194 mAb neutralizes SF162, VI 1888 and 92RW009 while the corresponding plasma do not produce significant neutralization against any isolate in these assays. The HK20 mAb only neutralizes VI 1888. When the absorption and culture phases of the assay are extended to the 1/24/14 setup, HJ16 still neutralizes VI 1090, HGN194 neutralizes SF162 and VI 1888 while HK20 neutralizes VI 1090 for 80%. With regard to cell line based assays, there is a good concordance for HGN194 mAb and plasma in the GHOST-PV assay. Four of these HGN194-PV combinations are also neutralizing when the target cells are TZMbl. However, when infectious virus is used 10/12 combinations are enhancing. HJ16 mAb neutralizes 92RW009 and VI 1090 in all three cell-line assays while the corresponding plasma failed to do so in these assays (see Table 5 ). Remarkably, however, 92RW009 was neutralized by 242315-HJ plasma in the 1/2/7 PBMC assay selectively and VI 1090 was neutralized by HJ plasma in both the 24/1/14 and the 1/2/7 PBMC assay. The HK20 mAb only neutralizes VI 882 in the GHOST_PV assay, SF162 in the TZMbl_PV assay and does not neutralize any infectious virus. There was no consistent statistically significant correlation between the levels of neutralization reached in the 24/1/14 PBMC assay and the others except where only a few isolates were actually neutralized or neutralization levels were low. In order to link the present and previous studies, plasma were tested in TZMbl assays against a sub-panel of PV included in supplementary table 2 of reference 7. Comparisons are presented in Table 8 . Again, there were anomalies with mAbs neutralizing isolates which were not neutralized by the corresponding plasma and vice versa. Similarly, the mAbs showed a reduced range of neutralization relative to their corresponding plasma. Plasma from the 242315-HJ patient is very effective against the three tier 1 strains and also against five out of 11 tier 2 strains. In contrast, the corresponding HJ16 mAb is not able to neutralize the tier 1 strains and although effective against six out of 11 tier 2 strains, these are not always the same isolates as neutralized by the plasma. Plasma from the 314994-HGN patient is able to neutralize all tier 1 strains as well as four tier 2 strains while the HGN194 mAb is also able to neutralize the tier 1 strains and three out of 11 tier 2 strains. However, again, these are not always the same strains that are neutralized by the plasma. Plasma from the HK patient is able to potently neutralize the three tier 1 strains but none of the tier 2 isolates while the HK20 mAb is effective against only one of the three tier 1 strains and one of the 11 tier 2 isolates. The proportion of isolates neutralized by an individual plasma was also markedly dependent on the panel of HIV isolates used. The patients' plasma were initially selected in 24/1/14 PBMC assays and the ITM panel of 14 primary infectious HIV-1 isolates (Table 4 ). In the smaller, modified panel of PV used in Antwerp (Table 5) The present study is based on an extensive program to employ naturally occurring broadly neutralizing Abs from HIV-infected patients as templates for immunogen design against A, C and CRF02 primary viruses. From the memory B cell interrogation of such patients many mAbs were generated, but only three of these (HJ16, HGN194 and HK20) showed interesting novel broad neutralizing capacity. Since the plasma and ensuing mAb were selected in different neutralization assays, we wanted to explore and understand the behaviour of these exceptional plasma and mAbs in various neutralization assays, based on PBMC or cell lines, using primary infectious viruses or non-replicating PV. A first observation was that the three patients, from whom the neutralizing mAbs were generated, showed an intermediate breadth of neutralization, preferentially neutralizing subtype A (HGN patient) or C (HJ patient) or A and C (HK patient), which did not correspond with the subtypes of their infection. Another remarkable observation is that they all showed a low viral load without treatment at the time of sampling. Only patient HGN had the profile of a viraemic controller, whereas HJ was a chronic progressor and patient HK was probably still in an early phase. In the last two years several groups have reported the discovery of new and promising Abs [17, 18, 19] . The HJ16, HGN194 and HK20 Abs obtained by our consortium were amongst those obtained by means of the interrogation of rather chronically HIV-1 infected patients. In the present study, the HGN194 patient was infected for at least 10 years and seemed to naturally control her HIV-1 infection, the HK20 patient may not have been infected for longer than a year and the HJ16 patient regularly required antiretroviral therapy to control her viral load. These data confirm that neutralizing Ab development does not protect against disease progression. Similarly, some of the broadest neutralizing plasma were obtained from patients who urgently required HAART (e.g. HY-plasma table 2, clinical history not shown). A side-by-side comparison of the different neutralization assays used for characterization of these patients' plasma and the newly isolated Abs showed that the broadest spectrum of strains and subtypes was neutralized in the PV assays as well as in the 24/1/ 14 extended incubation PBMC assay with primary virus. In contrast, the classical short incubation phase assays as well as the extended absorption phase PBMC assays showed a reduction in the number of neutralized strains. The TZMbl assay using primary virus also showed this restricted profile despite the fact that it has an extended absorption phase in common with the cell line PV assays. Results for the three patient plasma that were selected for their cross-neutralizing capacity in the 24/1/14 PBMC assay correlated with those obtained in the TZMbl_PV assay for only two patients. It is unusual that these two substantially different techniques result in comparable neutralization profiles (own results and [20] ). Both PV based assays correlated strongly with each other. Another observation was that all three isolated Abs have a narrower and partially different neutralization spectrum relative to the corresponding plasma in the extended incubation PBMC and TZMbl PV assays. Results with the HJ16 mAb from the PBMC, TZMbl and GHOST assays show good correspondence while for the HGN194 mAb the GHOST neutralization responses are broader. The HK20 mAb shows little to no neutralization in either the TZMbl or GHOST assays. Neutralization breadth across subtypes is unlikely to be due to endotoxin since plasma are negative in conventional assays where absorption phases (and therefore contact between plasma and cells) are longer [21] . Several factors may be responsible for the reduction in the range of isolates neutralized by the mAbs. One reason could be the polyclonal character of the Abs in the plasma. Cross-neutralization may require interaction between Abs acting at several epitopes. In this scenario, reproducing the range of isolates neutralized by plasma would not be possible when an average of only one to two neutralizing mAb were isolated. This would be buttressed by methods to directly determine the number of individual neutralizing antibody clones in the patient's repertoire. We will also address this issue in new studies but this has also been examined by the group of Guan and Lewis [22] . Obviously, combining more isolated mAb might correspond better to the plasma results when additive and synergistic effects between Abs could be unveiled. Unfortunately, we did not obtain more mAbs for the 242315-HJ and 529552-HK patients. Although we did obtain more mAbs from the 314994-HGN patient, none of these, except for HGN194, were neutralizing in either the HOS or TZMbl assays. Nevertheless, there may be 'missing Abs' as has been previously suggested by the groups of Guan and Nussenzweig [16, 22] . However, the 'non-neutralizing' mAbs may still be relevant in the wider context since they could have other effector mechanisms such as Ab-dependent cell-mediated virus inhibition (ADCVI) or Ab-dependent cellular cytotoxicity (ADCC) through their FccRs [23, 24, 25 ]. An alternative explanation for why relatively few neutralizing mAb were obtained is that the primary screen was binding to monomeric or trimeric envelope protein in ELISA and this procedure might not be optimal. In the Walker study [17] , it was shown that the most potent Abs did not bind in an ELISA and even Abs that did bind had a low neutralizing profile. The inference was that specific quaternary protein structures should be used in a primary screening. This issue is being addressed with the most recent samples under interrogation at IRB within our consortium. It should also be noted that the Abs in the plasma probably originate in the plasma cells of the bone marrow while the mAbs are isolated from memory B cells in the circulation. These two cell populations may not produce the same range of Abs. It should be possible to culture individual plasma cells, clone their heavy and light chain variable regions and identify the IgG or IgA repertoires produced. Selection of the patients from whom the mAbs were isolated was extremely assay dependent. The patients who gave the three interesting mAbs would not have been selected if any of the alternative assays had been used. The influence of target cells on neutralization has already been observed both by us and others [10, 20, 26, 27] . In particular, there is a three way interaction effect between the virus, antibody and target cells. Especially MPER specific Abs are more potent in PBMC based assays [4, 26, 27] . Since our data show that the HJ, HK and HGN patients are more potent in PBMC neutralization assays with an extended incubation phase it could be envisioned that these special patients could have a high proportion of gp41 specific Abs in their plasma. In the past, naturally occurring cross subtype neutralizing Abs have already been used as templates for immunogen design but in most of these cases patients were selected using either the classical short PBMC assay (1/2/7 format) or PV assays and our results clearly show that a different group of patients is selected by the extended incubation PBMC assay (24/1/14 format). Testing the patient plasma against primary strains is also more stringent since the molecularly cloned PV seem to be more easily neutralized. Hence, we believe that our selection procedure against the primary ITM panel provided us with patients that had more potent responses. A possible reason for any increased sensitivity of primary vs. pseudo viruses for identifying patients with potent neutralizing Abs could be the higher number of envelope glycoprotein spikes on the primary viruses relative to the PV [20, 28] . An alternative factor might be the density of (co)receptors on target cells, which has been implied by Corti et al who reported potent neutralization by HK20 in the HOS assays but almost no potency in TZMbl assays [2, 3] . Since HK20 recognizes an epitope in the gp41 region this could partially be explained by the high level of CCR5 expression on the TZMbl cells making it more difficult for anti-gp41 Abs to be effective [2, 29] . Also, the pathway employed by PV to enter TZMbl cells may be relevant so that HK20 could have been hindered by events following uptake into an endosome [30] . Since the non-replicating PV constructs could not be assessed in the primary target cells the recent development of molecularly cloned constructs in a Renilla replication competent backbone is certainly a step forward in the development of a standardized PBMC based neutralization assay to assess neutralization in primary cells [31, 32] . It remains elusive whether the HJ16, HGN194 and HK20 mAbs would have been obtained from other patients. HK20 like Abs have been detected through ELISA and although the neutralizing capacity of this fraction was not shown it still provides proof that a significant number of HIV-1 infected patients have responded to the gp41-HR1 region which is only briefly exposed [2, 3] . Even after almost 30 years of HIV research and the ongoing search for correlates of protection, there is still a critical need to determine how effective different types of antibody effector mechanisms can be in prevention of HIV-1 infection. Although many groups have tried to identify which neutralization assay can predict in vivo protection, this issue is still open to debate [33] . In several SIV and SHIV macaque studies neutralizing mAbs have correlated with protection [34, 35, 36, 37, 38, 39, 40, 41, 42, 43] , but there are also multiple counter examples [44, 45] . In this context, the most compelling demonstration that pre-existing Abs can be protective comes from passive immunization studies with either IgG or mAbs [28, 34, 38, 39, 40, 41, 42, 43, 46, 47, 48, 49] . The most recent study uses the HGN194 mAb against a SHIV strain containing an 'early' envelope and emphasizes the importance of potent neutralizing Abs that confer protection against a heterologous mucosal challenge [4] . The latter is highly significant since future vaccines will need to be effective against these relatively resistant early founder strains before infection is established in vivo [50] . Taken together our observations show that a single neutralizing mAb from each of the three patients does not reflect the major neutralizing spectrum of the patients' plasma and there is no apparent correlation of the mAbs targeting HIV strains belonging to the subtype of virus infecting the patient. It is quite evident that different neutralization assays yield different results and it is still unclear which one is most predictive or suited to obtain neutralizing mAbs. Nevertheless, the strategy used for selection of plasma (in an extended incubation PBMC assay) and selection of mAb (based on ELISA binding and neutralizing capacity in a HOS_PV assay) yielded interesting new mAbs. A better understanding of in vitro neutralization characterizations of patient plasma and Abs and will hopefully lead to more effective ways of discovering new Abs that ultimately can be used for HIV-1 immunogen design and subsequent vaccine development.

EBV-gp350 Confers B-Cell Tropism to Tailored Exosomes and Is a Neo-Antigen in Normal and Malignant B Cells—A New Option for the Treatment of B-CLL

gp350, the major envelope protein of Epstein-Barr-Virus, confers B-cell tropism to the virus by interacting with the B lineage marker CD21. Here we utilize gp350 to generate tailored exosomes with an identical tropism. These exosomes can be used for the targeted co-transfer of functional proteins to normal and malignant human B cells. We demonstrate here the co-transfer of functional CD154 protein on tailored gp350+ exosomes to malignant B blasts from patients with B chronic lymphocytic leukemia (B-CLL), rendering B blasts immunogenic to tumor-reactive autologous T cells. Intriguingly, engulfment of gp350+ exosomes by B-CLL cells and presentation of gp350-derived peptides also re-stimulated EBV-specific T cells and redirected the strong antiviral cellular immune response in patients to leukemic B cells. In essence, we show that gp350 alone confers B-cell tropism to exosomes and that these exosomes can be further engineered to simultaneously trigger virus- and tumor-specific immune responses. The simultaneous exploitation of gp350 as a tropism molecule for tailored exosomes and as a neo-antigen in malignant B cells provides a novel attractive strategy for immunotherapy of B-CLL and other B-cell malignancies.

Epstein-Barr virus (EBV) is an almost ubiquitous human gamma herpes virus that infects resting human B-lymphocytes, including B-CLL cells, with high efficacy [1, 2] . EBV's B-cell tropism is mainly due to gp350, the viral envelope glycoprotein that interacts with the cellular complement receptor 2 (CR2, CD21) [3] on B cells. In EBV seropositive individuals, gp350 mainly elicits CD4+ T-cell responses [4] . Exosomes are endosome-derived membrane vesicles, which are released by cells of diverse origin including dendritic cells, cancer cells [5] and EBV-infected B cells [6] . Exosomes bud from endosomal membranes and accumulate in multivesicular bodies, which eventually fuse with the cellular membrane and release the contained vesicles. Exosomes are rich in lipids and membrane proteins like MHC molecules, TNF-R and tetraspanins [5] but their specific composition depends on the cell of origin. Exosomes either fuse to the recipient cell membrane or are engulfed by phagocytic cells in such a way that exosome proteins are degraded and loaded onto MHC class II molecules [7] . Obviously, exosomes can deliver proteins as cargo in a very immunogenic manner so that they efficiently reactivate specific CD4+ T cell clones [8] . Hence, exosomes can induce strong and epitope-specific immune responses [9, 10] and can be used as an alternative to transfer strategies using gene vectors and as promising vaccines [11, 12] . Chronic lymphocytic leukemia of B-cell origin (B-CLL) is the most common adult leukemia in the Western hemisphere. B-CLL is considered as a prototypic disease undergoing immune evasion as the malignant cells lack important accessory and co-stimulatory molecules. Thus, despite their expression of high levels of surface MHC class I and II molecules, which presumably present tumorassociated antigenic epitopes, the leukemic cells tend to induce tumor-specific T-cell anergy. Typically, activated T cells from patients show a significantly reduced expression of CD40 ligand (CD154) or are completely CD154-negative [13] . As a consequence, T cells from B-CLL patients cannot activate cells through the CD40 receptor. This interaction, however, is essential for CD40 signaling and subsequent induction of other immune accessory molecules like CD80 and CD86, which increase the antigen-presenting capacity of normal and B-CLL cells. On the other hand, the EBV-specific cellular immunity is relatively intact in these patients [2] . To overcome the dysfunction of potentially tumor-reactive T cells from patients with B-CLL, several approaches have been developed relying on the stimulation of B-CLL cells through the CD40 pathway, including the ectopic expression of CD154 on the leukemic cells, and aiming at the selfstimulation of these cells [14] [15] [16] [17] . In summary, immunotherapy of B-CLL is promising and CD154 is a potential candidate molecule to improve the patients' immune status and, eventually, the clinical outcome. The robust cellular immunity in B-CLL patients against EBV [2] therefore prompted us to investigate the potential of tailored exosomes to redirect this immunity to malignant B cells. We present a novel approach for the targeted transfer of functional cellular proteins to B cells via tailored gp350+ exosomes. In this approach, gp350 has a dual function: (i) it confers B-cell tropism to exosomes so that they specifically co-transfer proteins of interest and (ii) it is a viral neo-antigen for these cells so that they efficiently reactive gp350-specific T cells. As a proof of concept, we show that tailored gp350+ exosomes can co-transfer functional CD154 as immune accessory molecule to B-CLL cells, which are subsequently stimulated to express surface molecules like CD54, CD80, CD86 and CD95 and stimulate autologous tumor-and EBVspecific T cells. EBV gp350 is packaged into exosomes, confers B-cell tropism, and reactivates specific T cells EBV has a profound B-cell tropism that is mainly conveyed by gp350, which is the major EBV glycoprotein in the viral envelope and the ligand for cellular CD21 (CR2) on B cells. We knew from previous work that exosomes can transport ectopically expressed proteins such as green fluorescent protein (GFP), which is presumably present as a cargo in the exosomal lumen. In addition, several groups provided evidence that surface proteins are incorporated in exosome membranes [5] . We therefore asked whether gp350 could also become an integral part of exosomes and confer B-cell tropism to these vesicles. To answer this question, we cotransfected 293 cells with expression plasmids encoding BLLF1, the gene of gp350, and gfp. Three days later, we isolated vesicles from the supernatants of transfected HEK293 cells as described in Material and Methods and analyzed them by immunoblots for the presence of gp350 and exosome markers. Gp350 was detected in vesicles that floated at a density between 1.03 and 1.08 into an OptiPrep TM gradient, corresponding to a density between 1.13 to 1.18 in a sucrose gradient and thus in the density described for exosomes. The gradient also revealed the co-sedimentation of gp350 with the exosome markers hps70, tsg101 and CD63, indicating the nature of the gp350+ vesicles as exosomes ( Figure 1A ). Flow cytometry of exosomes coupled to latex beads revealed that gp350 is presumably located within the exosome membrane because it could be targeted with a specific antibody ( Figure 1B ). To demonstrate that gp350 confers B-cell tropism also to exosomes, we incubated gp350+/gfp+ exosomes with PBMCs from a healthy donor for one day and then quantified exosome binding by measuring GFP fluorescence by flow cytometry. This assay revealed that gp350+/gfp+ exosomes had an EBV-like tropism because they bound to CD19+ B cells but not to CD19negative cells ( Figure 1C ). Phagocytic cells engulf exosomes, process their proteins in lysosomes and present epitops in association with MHC class II molecules to CD4+ T cells [10] . To further utilize the potential of gp350+ exosomes to specifically transfer exogenous proteins to B cells, which, in turn, may activate specific T cells, we generated exosomes that carried BNRF1, the major tegument protein of EBV, either alone (BNRF1+) or together with gp350 (BNRF1+/ gp350+). We then incubated purified CD19+ B cells with these exosomes overnight and used these PBMC as stimulators for an autologous BNRF1-specific CD4+ T-cell clone. As shown in Figure 1D , B cells incubated with BNRF1+/gp350+ exosomes activated the T-cell clone in a concentration-dependent manner whereas B cells incubated with BNRF1+ exosomes did not activate the T-cell clone. This result demonstrates the potential of gp350+ exosomes to transfer immunogenic foreign proteins to B cells that then can activate specific CD4+ T lymphocytes. In a next series of experiments we wanted to elucidate whether 293/gp350+ exosomes can co-transfer functional membrane proteins to B cells. As a model system but also as a potential practical application, we chose B-CLL cells that express CD21 and become immunogenic upon ectopic expression of CD154 on malignant cells. We, therefore, transfected 293 cells with expression plasmids for gp350 and CD154 and isolated exosomes as described above. Again, an immunoblot with a CD154-specific antibody demonstrated the presence of CD154 in exosome preparations and an OptiPrep TM gradient revealed co-sedimentation with gp350 ( Figure 1A) indicating the presence of both proteins on exosomes ( Figure 2A ). In addition, flow cytometry revealed that gp350+/CD154+ exosomes specifically bound to CD19+ B cells from a B-CLL patient ( Figure 2B ). To answer the question whether CD154 in gp350+/CD154+ exosomes was functional, HEK293 cells were transfected with a CD154 expression plasmid either alone or in combination with a gp350 expression plasmid. Transfer of exosomes to B-CLL cells was measured with an anti-CD154 antibody by flow cytometry. As shown in Figure 2C , exosomes that carry both proteins efficiently conveyed CD154 surface expression to B-CLL cells. CD154+/ gp350-exosomes were only transferred very inefficiently to B-CLL cells, probably due to an only weak interaction of CD154 with its receptor CD40 on the cell surface. In order to investigate the immune accessory function of CD154, we loaded B-CLL cells from an HLA II DR13+ donor either with 293 exosomes, with exosomes from 293 cells transfected with an expression plasmid for gp350 (gp350+ exo) or transfected with expression plasmids for gp350 and CD154 (CD154+/gp350+ exo). One day later, we added HLA-DR13-restricted gp350-specific CD4+ T cells and measured their activation with an IFN-c ELISA. This assay demonstrated the relevance of CD154 for T-cell recognition because B-CLL cells loaded with CD154+/gp350+ exosomes were significantly better stimulators than B-CLL cells loaded with gp350+ exosomes ( Figure 2D ). Induction of the accessory molecule ICAM-1 (CD54), the co-stimulatory molecules B7.1 (CD80) and B7.2 (CD86), and the death receptor Apo1 (CD95) on B-CLL cells upon incubation with CD154+/gp350+ exosomes ( Figure 2E ) but not with exosomes from 293 cells provided an explanation for improved recognition by T lymphocytes. We knew from previous experiments that PBMCs from healthy donors, which were incubated with gp350+ exosomes, efficiently re-stimulated an autologous gp350-specific CD4+ T-cell clone (data not shown). These experiments implicated that gp350, which is equally transferred to B-CLL cells by CD154+/gp350+ exosomes, can act as a neo-antigen in B-CLL cells, which are normally not infected with EBV [18] . This is a very interesting aspect because B-CLL patients usually maintain a robust and easily recruited CMV- [19] and EBV-specific T-cell response [2] although their T cells in general are functionally impaired. For instance, B-CLL cells infected with EBV or loaded with CMVpeptides are efficiently killed by autologous T lymphocytes from B-CLL patients [18, 20] . We thus aimed at investigating whether gp350 serves as a viral neo-antigen in exosome-treated leukemic cells and whether these cells become targets for gp350-specific autologous T lymphocytes. We stimulated PBMCs from B-CLL patients three times with CD154+/gp350+ exosomes as described above and elucidated the specificities and cytolytic potential of the activated T cells. To this end, we incubated autologous PBMCs for 24 hours with either CD154+/gp350+ exosomes, CD154+/gp350-exosomes or unmodified 293 exosomes or left them untreated. The next day, these cells were labeled with calcein and used as targets for autologous effector T cells. As shown in Figure 3 , T cells generated by stimulation with gp350+/ CD154+ exosomes efficiently killed autologous PBMCs that had been activated with CD154+/gp350-or CD154+/gp350+ exosomes as quantified by the release of calcein. Of interest, we did not observe notable activation of T cells against 293 proteins as PBMCs incubated with exosomes from non-transfected 293 cells were lysed to almost the same extent as non-activated autologous PBMCs. These experiments also demonstrated that PBMCs loaded with CD154+/gp350+ exosomes were constantly better lysed than PBMCs loaded with CD154+/gp350-exosomes. This prompted us to investigate whether T cells specific for gp350 accounted for this improved lysis. Therefore, we tested the cytolytic activity of these T cells against EBV-infected lymphoblastoid cells (LCLs), which are known to be targets for gp350-specific T cells [21] , and against EBV-negative B blasts [22] from an HLA-DR13+ matched healthy donor. We found that the T cells efficiently lysed allogenic LCLs, whereas they completely ignored EBV-free B blasts, indicating the presence of EBV-specific T cells in the effector population ( Figure 3 ). These data corroborate that the stimulation of PBMCs with CD154+/gp350+ carrying exosomes is an efficient method for the activation of B-CLL cells and the reactivation and Immunoblots demonstrated that vesicles carry gp350 and that these vesicles co-sediment with vesicle that carry the exosome markers hsp70, tsg101, CD63 and ganglioside (GM)1 (www.exocarta.org). Calnexin is present in cell lysates (CL), its absence from exosomes demonstrated the purity of the preparation. (B) Latex beads were coated with gp350+ exosomes and stained with a gp350-specific antibody. gp350 is accessible and thus probably located in the exosome membrane (isotype control is shown as grey tinted histogram). (C) gp350 confers B-cell tropism to exosomes. PBMCs from a healthy donor were incubated for 18 h with gp350+/gfp+ exosomes and then analyzed for GFP fluorescence by flow cytometry. Only CD19+ B-cells stained positive, whereas an interaction of gp350+ exosomes with CD3+ T cells and CD19-cells (T and NK cells, monocytes) could not be observed. An immunoblot with a CD154-specific antibody revealed that CD154 is highly expressed in transfected cell and that the protein is incorporated into exosomes. Shown are immunoblots of cell lysates (CL) and lysates from purified exosomes (EL) of normal and transfected 293 cells, incubated with CD154-(upper panel) and tsg101-specific antibodies (middle panel). An OptiPrep TM gradient revealed co-sedimentation of CD154-and gp350-carrying vesicles (see Figure 1A ). (B) Exosomes carrying gp350+ specifically bind to CD19+ B cells from a B-CLL patient. Fresh B-CLL cells were incubated with purified exosomes from 293 cells transfected with a gp350 expression plasmid and binding of exosomes was measured by flow cytometry. (C) Exosomes carrying CD154 bind only weakly to B-CLL cells, probably through interaction with CD40, which is highly expressed on these cells. Binding of CD154-carrying exosomes is drastically enhanced by the co-expression of gp350. 293 cells were co-transfected with expression plasmids for CD154 and/or gp350. Fresh B-CLL cells were co-cultivated with CD154+/gp350-and CD154+/gp350+ exosomes for two days and binding of exosomes and thus transfer of CD154 was measured by flow cytometry. Having demonstrated that B-CLL cells activated by CD154+/ gp350+ exosomes achieve an activated phenotype, we next wanted to assess whether these B-CLL cells became immunogenic to autologous EBV-and CLL-specific T cells. Given the expected low number of these T cells in peripheral blood, we stimulated 1,5610 7 PBMCs from patients with B-CLL three times within 21 days with lethally irradiated autologous PBMCs that were incubated with different exosomes as described above. One typical experiment is shown in Figure 4 : before the first stimulation on day 0, PBMCs of this patient consisted mainly of malignant B cells with only 4% CD3+ T cells. After the third round of stimulation, cultures incubated with gp350+/CD154+ exosomes almost exclusively contained CD4+ and CD8+ T-cells ( Figure 4A ). In total, stimulation with CD154+/gp350+ exosomes yielded approximately 1610 7 vital cells on day 31, meaning that the CD4+ and CD8+ T cell counts had increased about 25-fold and 15-fold, respectively ( Figure 4B) , while the total cell number slightly decreased ( Figure 4C ). In contrast, no viable cells were detectable in those cultures that were treated with CD154-/gp350exosomes derived from non-transfected 293 cells, or that were left untreated. Thus, stimulation of T cells with B-CLL cells loaded with CD154+/gp350+ exosomes is a powe rful option to selectively expand specific T cells from B-CLL patients in vitro. Next we wanted to elucidate the specificities and cytolytic activity of these T cells in more detail. As mentioned above, restimulated T cells form B-CLL patients efficiently lysed autologous PBMCs that were activated with CD154+/gp350-exosomes. These exosomes do not contain gp350 and, therefore, the T cells must have recognized cellular antigens. To test whether T cells specific for B-CLL-associated antigens had been reactivated, we incubated them with irradiated EBV-negative B-blasts [22] from an HLA-A2-matched donor as a negative control or B-blasts loaded with HLA-A2 restricted peptides derived from B-CLLassociated antigens, namely MDM [23] , ETV5 [24] and PU.1 [25] . As shown in Figure 4D , peptide-loaded B-blasts were efficiently lysed by the T cells stimulated with CD154+/gp350+ exosomes whereas B blasts not loaded with peptides were completely ignored. Lysis of target cells was efficiently reduced upon addition of an MHC class I-specific antibody, W6/32, demonstrating the antigen specificity of the effector T cells. CD154 is a known promising candidate molecule for the immunotherapy of B-CLL because it increases the immunogenicity of the tumor cells. In line, the ectopic expression of CD154 on B-CLL cells was shown to induce the expression of important costimulatory and adhesion molecules on the leukemic cells, turning them into efficient stimulators of autologous T cells. In principle, viral gene transfer of CD154 into B-CLL cells is a suitable approach to induce immunological reactions both in vitro and in vivo [14, [26] [27] [28] [29] but the low transduction efficiency and the resulting high dose of viral vectors may induce severe side effects. There also remain major concerns about the clinical use of viral vectors. Exosomes are cellular microvesicles, which have already demonstrated their potential to induce specific immune responses. The majority of permanent cell lines spontaneously release exosomes into the supernatant, from where they can be easily purified and concentrated [30] . Transfecting HEK293 cells with expression plasmids coding for CD154 and gp350 resulted in the secretion of modified exosomes carrying both proteins. Upon interaction with their target B cells, an undetermined fraction of exosomes is presumably engulfed and degraded in lysosomes in such way that peptides of exosomal proteins are presented in association with MHC class II molecules. Exosomes that are not taken up by the target cells probably engage in protein-protein contacts with cellular surface molecules and lead to receptor activation and signaling. gp350 molecules on the surface of exosomes probably target these vesicles exclusively to human B cells which express relevant amounts of the receptor molecule for gp350, CD21. Exosomal co-transfer of both gp350 and CD154 is thought to lead to efficient CD154/CD40 ligand/receptor engagement on the B-cell surface activating the intrinsic CD40 signaling cascade. Treatment of B-CLL cells with exosomes transferring functional CD154 protein causes leukemia cells to become efficient APCs. Activating the CD40 receptor leads to the induction of immune accessory molecules, making these leukemic cells potent stimulators of autologous T lymphocytes. Since almost all patients with CLL have a high prevalence of EBV (94%) [31] and thus possess EBV-specific memory cells with a high frequency of CD4+ gp350specific T cells [32] , the incorporation of gp350 into exosomes has a dual function: it confers B-cell tropism and serves as an immunodominant viral CD4 antigen. The incorporation of a viral protein as a tumor-specific antigen is straightforward, because the cellular immune system of cancer patients is usually impaired but virus-specific immune responses are detectable even in late-stage patients and T lymphocytes specific for herpes viruses are often present at high numbers [1, 33] . Given the fact that the vast majority of CLL patients are seropositive for EBV, gp350 and other EBV proteins have the potential as promising neo-antigens in B-CLL cells exploiting and redirecting the strong anti-viral cellular immune responses to leukemic cells. Based on these results, we propose here a new immunotherapeutic approach for B-CLL, based on the simultaneous targeted transfer of functional CD154 and the EBV protein gp350 onto malignant cells using exosomes. gp350 is the major envelope protein of EBV and confers viral B-cell tropism by interacting with the complement receptor 2 (CD21), which is highly expressed on B lymphocytes. Here, we generated modified exosomes produced in 293 cells. As a result of gp350 incorporation, these particles have a profound B-cell tropism similar to wild-type EBV, so that gp350carrying vesicles specifically and efficiently bind to B cells. In addition, gp350 serves as a viral neo-antigen in B-CLL cells. We also found that CD154 on exosomes from HEK293 cells is functionally active as demonstrated by the induction of immune accessory molecules on B target cells probably through the CD40 pathway. Taken together, our experiments suggest that leukemia cells treated with CD154+/gp350+ exosomes are efficiently stimulated and subsequently killed by autologous B-CLL and gp350-specific cytolytic T lymphocytes. In summary, our results demonstrate that modified exosomes carrying the EBV protein gp350 display a distinct tropism to normal and leukemic B cells and efficiently transfer CD154 as a functional protein onto these cells. Leukemic B cells treated with these particles acquire an activated phenotype and become potent stimulators of autologous T lymphocytes. Engineered exosomes can be easily generated and can readily be scaled up for clinical applications. In addition, they can be individually tailored to express additional accessory molecules like OX40L or the Fas ligand or alternative viral molecules to target other classes of cells like macrophages and DCs. The generation of modified exosomes is not limited to 293 cells, which we used in this proof-of-concept. Instead, other cell lines that are approved for human therapy, such as MRC-5 fibroblasts, should also be tested as an optional origin of exosomes to facilitate transition into clinical trials. Modified gp350-carrying exosomes can thus be regarded as powerful and promising tools for various immunotherapeutic approaches. Peripheral blood samples were obtained from patients with diagnosis of B-CLL after informed consent approved by the Institutional Ethics Committee. Normal blood samples were taken from healthy volunteers. Mononuclear cells were isolated by density gradient centrifugation on F/H. B cells were purified with CD19-specific MACS beads (Miltenyi, Bergisch Gladbach, Germany). Cells were cultured at 37uC in a 5% CO 2 atmosphere in standard medium with 10% fetal calf serum. HEK293 is a human embryonic kidney cell line [34] , which spontaneously releases exosomes into the cell culture supernatant. For the generation of modified exosomes, 293 cells were cotransfected with expression plasmids for BLLF1 (gp350), CD154 and/or gfp. For the isolation of exosomes, 25 ml of conditioned supernatants were collected three days later, sterile filtrated and subjected to repeated centrifugations at increasing centrifugal force (10 min at 300 x g, 10 min at 5000 x g in a Heraeus 3SR+centrifuge, followed by 2 h at 100.000 x g in a Beckman LE-80K ultracentrifuge in a SW28 swing-out rotor. The pelleted particles were washed, resuspended in 500 ml volume PBS containing protease inhibitors (Complete Mini, Roche) and the protein content was analyzed in a Lowry microassay using reagents purchased from Bio-Rad (Munich, Germany). Exosomes were further purified by flotation into a 5-30% iodixanol gradient (OptiPrep TM , Sigma Aldrich, Deisenhofen, Germany). B-CLL cells were cultivated with 100 mg of exosomes in a final volume of 2 ml for two days. Induction of surface accessory molecules was measured by flow cytometry using a FACS Calibur flow cytometer (Becton Dickinson). For bead-coupling assays 5 ml of surfactant-free sulfate/aldehyde latex beads were incubated with 10 mg of exosomes for 15 min at room temperature. Then 1 ml of PBS was added and the beads were incubated for another 2 hours. The beads were washed three times in PBS with 2% FCS and analyzed. Antibodies specific for tsg101 (sc-978) and CD154 (sc-7964) were purchased from Santa Cruz Biotechnology. E. Kremmer (Munich) provided the gp350-specific antibody 72A1. Fluorochrome-labeled secondary antibodies were obtained from Becton Dickinson (Heidelberg, Germany) or Immunotools (Friesoyte, Germany). Vesicle preparations were spotted onto a PVDF membrane, incubated with with specific primary antibodies and an HRPcoupled secondary antibody and developed with the ECL system (GE Healthcare). Ganglioside M1 was detected with colera toxin (Sigma Aldrich). In order to reactivate EBV-specific T-cells in B-CLL blood samples, 3610 7 cells were cultivated with 100 mg of exosomes in a final volume of 5 ml. Re-stimulations were performed on days 14 and 28 by adding 1610 7 lethally irradiated autologous PBMCs that have been loaded with 100 mg exosomes for 5h. Fresh medium was added once a week. After 31 days, cells were analyzed by flow cytometry for lineage markers. Interferon-cELISA assays were performed according to the manufacturers instructions (Mabtech, Uppsala, Sweden). The gp350-and BNRF1-specific CD4+ HLA-DR13-restricted T-cell clones used have been described elsewhere [8, 35] . All cells were tested in triplicates. Calcein release assays were performed as described previously [36] . Briefly, target cells were labeled with Calcein-AM (1% solution in medium) for 30 min at 37uC. After intensive washing the cells were either incubated with 10 mg of exosomes for 4 h or loaded with CLL-specific, HLA-A2-restricted peptides (PU1.423: aa-sequence VLFYLGQYI, mdm2.53: YLAPENGYL and ETV5.45: ELFQDL-SQL) at 20 mg/ml for 1 h or left untreated. The cytotoxicity assays were performed over a time period of 6 h, using an effector-targetratio of 40:1, and the calcein released into the supernatant was quantified with excitation and emission wavelengths of 490 nm and 530 nm, respectively, in a plate-reader (Perkin Elmer, Waltham). All experiments were performed at least three times.

Detection of a Fourth Orbivirus Non-Structural Protein

The genus Orbivirus includes both insect and tick-borne viruses. The orbivirus genome, composed of 10 segments of dsRNA, encodes 7 structural proteins (VP1–VP7) and 3 non-structural proteins (NS1–NS3). An open reading frame (ORF) that spans almost the entire length of genome segment-9 (Seg-9) encodes VP6 (the viral helicase). However, bioinformatic analysis recently identified an overlapping ORF (ORFX) in Seg-9. We show that ORFX encodes a new non-structural protein, identified here as NS4. Western blotting and confocal fluorescence microscopy, using antibodies raised against recombinant NS4 from Bluetongue virus (BTV, which is insect-borne), or Great Island virus (GIV, which is tick-borne), demonstrate that these proteins are synthesised in BTV or GIV infected mammalian cells, respectively. BTV NS4 is also expressed in Culicoides insect cells. NS4 forms aggregates throughout the cytoplasm as well as in the nucleus, consistent with identification of nuclear localisation signals within the NS4 sequence. Bioinformatic analyses indicate that NS4 contains coiled-coils, is related to proteins that bind nucleic acids, or are associated with membranes and shows similarities to nucleolar protein UTP20 (a processome subunit). Recombinant NS4 of GIV protects dsRNA from degradation by endoribonucleases of the RNAse III family, indicating that it interacts with dsRNA. However, BTV NS4, which is only half the putative size of the GIV NS4, did not protect dsRNA from RNAse III cleavage. NS4 of both GIV and BTV protect DNA from degradation by DNAse. NS4 was found to associate with lipid droplets in cells infected with BTV or GIV or transfected with a plasmid expressing NS4.

The genus Orbivirus currently includes twenty two distinct virus species, with genomes composed of 10 segments of linear double stranded RNA (dsRNA), that are vectored by Culicoides midges, ticks, phlebotomine flies, anopheline or culicine mosquitoes. The three economically most important orbiviruses: Bluetongue virus (BTV) (the Orbivirus 'type-species') African horse sickness virus (AHSV) and Epizootic hemorrhagic disease virus (EHDV) are all transmitted by Culicoides biting-midges [1] . Several tick-borne orbiviruses can infect humans, including members of the Changuinola virus, Corriparta virus, Lebombo virus, Orungo virus and Great island virus (GIV) species. The coding assignments of the 10 BTV genome segments were initially determined in 1983 [2, 3, 4] . Seven distinct structural proteins (VP1 to VP7) and 3 distinct non-structural proteins (NS1, NS2 and NS3) were identified in orbivirus infected cells, or after in vitro translation of viral RNA. In most cases each genome segment encodes a single protein from a single open reading frame (ORF), expect segment 9 (Seg-9) and segment 10 (Seg-10), both of which encode two nearly identical proteins initiated from in-phase AUG codons close together near the upstream termini (VP6 and VP6a encoded by Seg-9, and NS3 and NS3a encoded by Seg-10) [2, 5] . However, in vitro translation of BTV RNA segments reproducibly generated a number of smaller translation products of unknown significance, that were usually dismissed as unimportant byproducts of translation [2] . The icosahedral orbivirus core-particle is constructed as two concentric protein shells, the sub-core layer which contain 120 copies/particle of the T2 protein (VP3 of BTV), and the coresurface layer composed of 780 copies/particle of the T13 protein (VP7 of BTV). VP1, VP4 and VP6 are minor enzymatic proteins that are packaged along with the ten genome segments within the central space of the virus core [6, 7] . The orbivirus outer-capsid layer is composed of two additional structural proteins (VP2 and VP5 of BTV), which mediate cell-attachment and penetration during initiation of infection. These outer-capsid proteins are more variable than the core proteins and most of the non-structural proteins, and the specificity of their reactions with neutralising antibodies determines the virus serotype (as exemplified by VP2 of BTV [8] ). The relative number and locations of the BTV structural proteins have been determined in biochemical and structural studies using cryo-electron microscopy and X-ray crystallography [7, 9, 10, 11, 12] . NS1 is the most abundant protein in BTV infected cells, forming tubules that may be involved in translocation of progeny virus particles to the cell membrane [13, 14] . BTV NS2 can be phosphorylated by ubiquitous cellular kinases and is an important matrix protein of the granular viral inclusion bodies (VIB) that form within the cytoplasm of infected cells. VIB represent the primary site of virus replication and assembly. The smallest of the BTV non-structural proteins that were previously identified, are membrane glycoproteins NS3 and NS3a, which are expressed in large amounts in insect cells, but not in mammalian cells. They are involved in the release of progeny virus particles from infected cells [15] . In some orbiviruses (e.g. AHSV) NS3/NS3a are highly variable and it has been suggested that they may be involved in determination of both vector competence and virulence [16] . BTV-Seg-9 encodes the minor core protein VP6, which is a helicase. Recent bioinformatic analyses have identified a new overlapping ORF in Seg-9 of both insect-borne and tick-borne orbiviruses, although the putative protein (identified here as NS4) varies in size between 10 kDa and 22.5 kDa [17, 18] . We report the synthesis and detection of NS4, in the cytoplasm and nuclei of cells infected with insect-borne and tick-borne orbiviruses (represented by BTV and GIV respectively). All animal immunisation work was conducted according to the recommendations in the Animals (Scientific procedures) Act of the Home Office of the UK and the Directive on the protection of Animals used for Experimental and other scientific purposes of the EU. The protocol was approved by the Ethics Committee of animal experiments at the Institute for Animal Health in the UK (Project license number 70/7060). All surgery was performed under sodium pentobarbital anaesthesia, and all efforts were made to minimize suffering. BHK-21 (American type cell culture collection) were grown at 37uC under 5% CO2 in Glasgow's minimum essential medium (GMEM), supplemented with 10% foetal bovine serum, 10% tryptose phosphate broth, penicillin G (100 IU/ml) and streptomycin (100 mg/ml). Culicoides sonorensis KC cells were grown at 28uC in Schneider's insect medium supplemented with 15% fetal bovine serum. Confluent monolayers of BHK-21 cells were infected with either BTV-8 (isolate NET2006/04) or Great Island virus (GIV) (isolate CAN1971/01) at a multiplicity of infection (MOI) of 0.1 pfu/cell. Infected cell cultures were incubated at 37uC for 72 hours until cell lysis began. The cells were then scraped into the supernatant and centrifuged at 3,000 g for 10 minutes. RNA was extracted from cell pellets using guanidinium isothiocyanate (RNA NOW reagent: Biogentex, Tx, USA) as described earlier [19] . KC cells were infected at an MOI of 0.1 pfu/cell and then incubated at 28uC for 7 days. Both BHK-21 and KC cell pellets were used in western blot analyses as described below. Viruses were purified from BHK-21 infected cells, as previously described using a discontinuous sucrose gradient [20] . Virus particles formed a blue opalescent band at the interface of the sucrose solutions. This was recovered and further purified by layering onto a continuous PercollH gradient as previously described [21] , using an SW41 rotor (100000 g, 1 hour, 4uC). The virus formed a blue band which was collected, diluted in 0.1 M Tris-HCl and pelleted at 10000 g for 1 hour. Bioinformatic analyses of the overlapping ORF in Seg-9 of BTV and GIV The hydrophobicity profile of different NS4 proteins was analysed using the Kyte and Doolittle hydrophobicity plot with a window size of 11 amino acids (aa) [22] . Sequence relatedness to proteins in public databases was assessed using the NCBI's BLAST (http://blast.ncbi.nlm.nih.gov/Blast.cgi)) and the pfam software (http://pfam.sanger.ac.uk/search/sequence). Amino acid alignments of NS4 of various orbiviruses were generated using the Clustal X program [23] and pairwise aa identities calculated using the MEGA 4 package [24] . The presence of 'coiled-coils' was indicated by analyses using the program 'COILS' (http://www.ch. embnet.org/cgi-bin/COILS_form_parser) and the PredictProtein server (http://www.predictprotein.org). The presence of nuclear localisation signals were analysed by PredictNSL, implemented in the PredictProtein server, and the cNLS Mapper (http://nlsmapper.iab.keio.ac.jp/cgi-bin/NLS_Mapper_form.cgi). Synonymous site conservation within the BTV VP6 coding sequence was analysed as described previously [25] . For this procedure, alignment columns in which the reference sequence (GenBank accession number: NC_006008) contained gap characters were removed so that the plots are in reference sequence coordinates. The RNA of BTV-8 or GIV was separated by 1% agarose gel electrophoresis. Seg-9 was cut from the gel using a clean scalpel blade, purified using RNaid kit (MP Biomedicals) and cDNA was synthesised using a single primer amplification technique as previously described [19] ). The ORFs in Seg-9 from BTV-8 (between nucleotides 182 and 415, accession number: AM498059) and GIV (between nucleotides 176 and 748: accession number HM543473) were PCR amplified using specific primers tailed with restriction enzyme sites shown in table 1. The pGEX-4T-2 vector and Seg-9 PCR products were doubledigested with EcoRI and NotI (BTV-8) or EcoRI and XhoI (GIV) enzymes (Invitrogen). The pCI-neo vector and Seg-9 PCR products were double-digested with EcoRI and NotI (BTV-8) or EcoRI and XbaI (GIV) enzymes. Digested products were gel purified using Genclean kit (Qbiogen). Corresponding vectors and PCR products were ligated overnight (O/N) at 16uC using T4 DNA ligase (Roche) to generate pGEX-BTVNS4, pGEX-GIVNS4, pCI-BTVNS4 or pCI-GIVNS4. These recombinant plasmids were used to transform XL1-Blue bacteria (Stratagene). Clones were recovered and grown in trypticase-soy-casein (TSC) medium containing 100 mg/ml ampicillin. The plamsids were subsequently purified using Qiaquick plasmid miniprep kit (Qiagen) and sequenced using the D-Rhodamine DNA sequencing kit and an ABI prism 377 sequence analyser (Perkin Elmer). Confirmed pGEX-BTVNS4 or pGEX-GIVNS4 plasmids were used to transform BL21 or C41 bacteria. A single colony of each plasmid was grown overnight (ON) in TSC/ampicillin, then used to seed 200 ml of fresh TSC/ampicillin. The bacteria were grown until OD600 0.5, then 0.5 mM IPTG was added for induction, for 4 hours at 37uC, or for 8 hours at 28uC. The bacterial cells were pelleted and processed using Bugbuster protein purification (Novagen) as previously described [26] . The soluble fraction of the fusion protein was purified by glutathione affinity chromatography using glutathione sepharose, as directed by the manufacturer (GE Healthcare). Proteins were analysed by sodium dodecyl sulfate/polyacrylamide gel electrophoresis (SDS-PAGE) using a 10% polyacrylamide separating gel (Miniprotean III) with a 3% stacking gel and stained with Coomassie brilliant blue, as described Table 2 . Percentage amino acid identity values between NS4 of BTV, EHDV, AHSV, GIV, PHSV and YUOV. Amino acid identity ranged from 5% to50%. The highest identity exists between BTV and EHDV (50%) followed by PHSV and YUOV-1 (30%). Amino acid identity in NS4 between the tick-borne and insect-borne viruses ranged between 5% and 18%. doi:10.1371/journal.pone.0025697.t002 previously [21] . The purified fusion protein was used to immunize rabbits (Harlan) with an initial injection, followed by 4 boosts at 2 weeks interval in the presence of Montanide ISA50 (Seppic) as an adjuvant. BTV-8 or GIV infected BHK-21 cells (5610 6 cells) and BTV-8 infected KC cells (5610 6 cells) were dissolved for 10 min at 100uC in 1 ml of sample denaturation buffer (160 mM Tris-HCl, 4 mM EDTA, 3.6% SDS, 60 mM DTT, 0.2% ß-mercaptoethanol, 0.8% methionine, 800 mM sucrose). A volume of 20 ml was analysed per well, by electrophoresis in a minigel (Miniprotean III tank -Bio-Rad). Purified and pelleted virus particles were also dissolved in sample buffer and analysed by SDS-PAGE, using a 4-20% gradient polyacrylamide gel. Resolved proteins were electro-blotted on 0.2 mm nitrocellulose membrane (Bio-Rad) using 20 mM Tris, 0.05% SDS, 150 mM glycine and 20% V/V isopropanol transfer buffer. Membranes were blocked with 5% skimmed milk, in Tris buffered saline (TBS: 25 mM Tris/HCl, 150 mM NaCl, 2 mM KCl, pH 7.4) and incubated over night with a dilution of 1/300 rabbit antisera. Membranes were washed three times with TBS-Tween-20 (TBS containing 0.05% Tween-20) and further incubated with monoclonal, anti-rabbit, peroxydase conjugate (Sigma), diluted at 1/750 in 5% skimmed milk. After 2 hours the membrane was washed three times with TBS-Tween-20 and developed using 4-chloronaphthol (Sigma) in presence of hydrogen peroxide. Logarithmically growing BHK-21 cells were infected with BTV-8 (1pfu/cell) for 24 hours then harvested and washed once with PBS. Nuclear extracts were prepared from 2.5610 7 cells using the NE-PER nuclear and cytoplasmic extraction reagent kit (Pierce), as directed by the manufacturer. The nuclear extract was mixed volume to volume with sample denaturation buffer and analysed by SDS-PAGE using a 4-20% gradient polyacrylamide gel. Resolved proteins were electro-blotted on 0.2 mm nitrocellulose membrane as described above, blocked with 5% skimmed milk, in TBS-Tween-20 and incubated over night with a dilution of 1/300 anti-BTV-8 NS4 rabbit antiserum. The membranes were washed three times with TBS-Tween-20 and further incubated with monoclonal, anti-rabbit, peroxydase conjugate (Sigma), diluted at 1/750 in 5% skimmed milk. After 2 hours the membranes were washed three times with TBS-Tween-20 then incubated with Lumilight plus (Roche) chemiluminescent detection reagent, as described by the manufacturer. X-Omat radiographic films (Kodak) were exposed for 10 minutes to membranes then developed as described by the manufacturer. BHK-21 cells were grown on coverslips placed at the bottom of a 24 well plates. 50% confluent cells were infected with 0.1 pfu/ cell of BTV-8 or GIV, incubated at 37uC for 4 hours or 24 to 72 hours, then fixed in 4% paraformaldehyde and processed for immuno-fluorescence. Briefly, rabbit antisera raised against NS4 of BTV-8 or GIV and a mouse anti-alpha tubulin antibody were both diluted 1/500 in PBS containing 0.5% bovine serum albumin (PBS-A) and applied to the fixed cell. After 1 hour incubation at room temperature (RT), slides were washed in PBS, then incubated with Alexa Fluor 488 conjugated anti-rabbit IgG (Invitrogen) and Alexa Fluor 568 conjugated anti-mouse, both diluted 1/250 in PBS. After labelling with primary and secondary antibodies, the cells were stained with DAPI (1:10,000) for 15 BHK-21 cells grown in 24 well plates (75% confluence), were transfected in triplicate, with pCI-BTVNS4 or pCI-GIVNS4 (4 mg/well) using Fugene-6 (Roche). At 48 hours post-transfection, the cells were fixed in 4% paraformaldehyde and processed for immuno-fluorescence, using anti-NS4 antibodies, as described above. Identification of NS4 in cells transfected with pCI-BTVNS4, using anti-BTV-8 immune serum from infected mice BHK-21 cells were transfected with pCI-BTVNS4 using Fugene-6. At 48 hours post-transfection, the cells were dissolved in sample denaturation buffer as described above. Cell lysates were analysed by SDS-PAGE/Western blot, using an immune serum (diluted 1/50, in 5% skimmed milk) from mice infected with BTV-8. Non-transfected cells were used as control. Nucleic acid protection assays dsRNA binding proteins can compete with Dicer (an endoribonuclease of the RNAse III family), reducing its ability to cleave long dsRNAs into 21 bp-long 'interfering' RNAs [27] . A dsRNA ladder (New England Biolabs) with sizes ranging from 500 to 21 bp was used as a template for Dicer cleavage. A competition assay with Dicer (Mobitech), was carried in the presence of 150 ng of expressed BTV-8 or GIV NS4, in a final volume of 20 ml containing 8 ml of the Dicer reaction, 2 mg of dsRNA, 1 mM ATP, 2.5 mM MgCl2. The reaction was incubated at room temperature for 20 minutes, followed by addition of a 1.5 U of Dicer, then incubated for a further 6 hours at 37uC. The reaction products were analysed by 3% agarose gel electrophoresis. DNAse I is an endodeoxyribonuclease that can degrade dsDNA into 59 phosphorylated tetranucleotides [28] . A dsDNA ladder (Promega) with sizes ranging from 2645 to 36 bp provides a target for DNAse I cleavage. Competition assays between DNAse I (Roche) and BTV-8 or GIV NS4, were carried out in a final volume of 20 ml, containing 2 ml of 10X DNAse I buffer and 2 mg of dsDNA. The reaction was incubated at room temperature for 20 minutes, followed by addition of a 2 U of DNAse I, then incubated for a further 30 minutes at 37uC. The completed reaction was heated at 99uC for 1 minute to inactivate the DNase and the reaction products were analysed by 2% agarose gel electrophoresis. The outer capsid protein VP9 of Banna virus (BAV, genus Sedaornavirus, family Reoviridae) expressed in E.coli [29] was used as a control in both RNAse and DNAse assays. A colorimetric assay was developed to detect interactions between NS4 and dsRNA. Synthetic dsRNA was prepared with the 59-end of one strand linked to biotin via a 15-atom mixed polarity tetraethylene glycol spacer (59-Biotine TEG). This design allows the dsRNA to be captured at the bottom of a well of 96 well plate coated with streptavidin, while keeping the dsRNA free as a target for NS4 binding. The sequence of the +ve strand is: 59-Biotine-TEG-UGGAAGCGGCUGGCAAUUAAUUUUGGU-GUC-39 and that of the negative strand is 59-GACAC-CAAAAUUAAUUGCCAGCCGCUUCCA-39. Increasing concentrations (from 1 to 640 ng) of the dsRNA in PBS were added to separate wells of a streptavidin-coated 96 well plate (Pierce) and allowed to bind at room temperature for 2 hours. The wells were washed three times with TBS-Tween-20, then two hundred microlitres of a 5% solution of bovine serum albumin (BSA) in PBS, was added in each well, to block non-specific sites. After washing 3 times with TBS-Tween-20 a fixed amount (150 ng) of either BTV or GIV NS4 in binding buffer (20 mM Tris-HCl pH 7.5, 50 mM KCl, 2 mM MgCl2, 2 mM MnCl2 and 5% glycerol) was added per well, prior to incubation for 30 minutes at 25uC. After the wells had been washed 3 times with TBS-Tween-20, rabbit anti-BTV or anti-GIV NS4 sera was diluted 1/250 in 5% BSA and 100 ml was added to each well, then the plates were incubated at 25uC for 2 hours. After washing three times, 100 ml of peroxydase conjugated antirabbit antibody was added (diluted 1/750 in 5% BSA) to each well. The plates were incubated at 25uC for 2 hours, then washed 3 times with TBS-Tween-20. One hundred microliters of SureBlue TMB 1-component microwell peroxidase substrate (tetramethyl benzidine from KPL) was added per well, then incubated for 30 minutes at 25uC. The reaction was stopped by adding 100 ml of 1 M HCl and the plate was read at OD 450 nm. Wells not containing dsRNA/NS4, were included as negative controls. Wells from which the dsRNA was omitted, but in which NS4 (BTV or GIV) alone was incubated were also included as controls. The program MLOGD models and compares sequence evolution in single-coding and dual-coding sequences. It has previously been used to identify a second ORF, in a different but overlapping reading frame from that encoding the viral helicase (VP6 of BTV), within Seg-9 of the insect-borne orbiviruses [17, 18, 30] . This ORF was also identified in tick-borne orbiviruses [18] . The length of the putative translation product is highly variable, even between closely related Orbivirus species. In BTV and EHDV it is approximately 10 kDa, in Peruvian horsesickness virus (PHSV) and Yunnan orbivirus (YUOV) it is approximately 13.5 kDa, while in AHSV it is approximately 17 kDa, and in GIV it is approximately 22.5 kDa (twice as long as in BTV). These NS4 sequences contain a high proportion of charged residues, with basic R+K (arginine + lysine) content ranging from 13% to 22%, while acidic E+D (glutamic + aspartic acids) content ranges from 12% to 22%. Each NS4 protein contains 4-5 histidine residues, with the exception of the BTV protein, which contains none. The levels of pairwise nucleotide conservation at synonymous sites within aligned sequences of the VP6 ORF, were used to assess the functional importance of the NS4-ORF. Complete or near-complete VP6-encoding sequences from BTV showed strikingly enhanced conservation in the region corresponding to the NS4 ORF (figure 1), supporting and extending previous computational analyses [17] . Enhanced conservation was also apparent at the 59 end of the VP6 coding sequence, indicating that this region (like the terminal noncoding regions of orbiviruses) is likely to contain functionally important elements. Amino acid identity between NS4 of the different Orbivirus species compared ranged from 5% to 50%. Highest identity was detected between BTV and EHDV (50%), followed by PHSV and YUOV-1 (30%). Amino acid identity in NS4 between the tickborne and insect-borne viruses, ranged between 5% to 18% (table 2). Local blast analyses using BLAST-P or TBLAST-N identified significant matches (as defined by the E value in BLAST) between NS4 proteins encoded by other orbiviruses. Analysis of NS4 protein sequences using the pfam program, which uses the hidden Markov model (HMM) based profiles to identify or predict protein functionalities [31, 32] , revealed strong similarities to certain conserved functional motifs. AHSV NS4 exhibits strong relatedness over almost its entire length with DUF domains that have helical structures known to be involved in nucleic acid binding and/or modification [33] . Previous analysis of GIV NS4 identified a 72 amino acid fragment (aa 82 to 153) with 39% similarity to dsRNA-binding domains of similar length (approximately 68aa) in other reovirus proteins [ [18] or other dsRNA binding proteins [34] . BTV NS4 (77aa long) also exhibits relatedness (over aa 14-54) to a DUF domain, belonging to the MetJ/Arc repressor superfamily [35] , which has a ribbon-ribbonhelix-helix DNA-binding motif, with the beta-ribbon located in and recognising the major groove of operator DNA. BTV NS4 shows strong relatedness to fzo-mitofusin protein, a putative transmembrane GTPase. The fzo protein has a coiled-coil structure and mediates mitochondrial fusion [36] . Another protein family with a coiled-coil structure, which also shows a strong match with BTV NS4, is EMP24_GP25L. Members of this family have been implicated in transporting 'cargo' from the endoplasmic reticulum (ER) and are related to the previously described GOLD domain [37] , which is always found combined with lipid-or membrane-association domains. Sequence analyses indicate that PHSV NS4 (111 aa long) contains a coiled-coil domain between aa 75 and 111, YUOV NS4 (113 aa long) contains two coiled-coils domains between aa 5 to 45 and 75 to 105, and AHSV NS4 (143 aa long) contains coiled-coil domains between aa 5 to 85 and aa 110-140. The BTV NS4 (77 aa long) appears to contain only a single coiled coil structure, between aa 27 and 77. Two overlapping potential nuclear localisation signals' (NLS) were identified in the aa sequence of PHSV NS4 ( . Although all of these NLS were monopartite, the GIV NS4 was found to contain a bipartite NLS (position 113-141: RKRGLEFLLLPLHEYVTHCAKEDIR-IYES). The prediction cut-off scores for all these NLS as defined by PredictNSL and cNLS ranged from 4 to 8, indicating dual nuclear/cytoplasmic localisations of a given protein [38] . The aa region 55 to 129 of GIV NS4 showed 29% identity (55% similarity) to aa 1823 to 1890 of UTP20 (a component of the nucleolus). NS4 of BTV and GIV were successfully cloned into pGEX-4T-2 and expressed in C41 at 28uC, as partially soluble proteins fused to GST (figure 2). The soluble fraction was used in competition assays with DNase I or endoribonucleases belonging to the RNAse III family and in binding assays with dsRNA. In contrast when these proteins were expressed in BL-21 they were totally insoluble and formed inclusion bodies. The inclusion bodies fraction was purified using bugbuster reagent, solubilised and used for immunization of rabbits. Western blots analyses, using rabbit antisera raised against recombinant BTV-8 NS4, showed that NS4 is expressed in BTV-8 infected Culicoides KC cells and BHK-21 ( figure 3, figure 4) . The antiserum identified a single protein band with an apparent molecular weight of approximately 12 kDa in BTV-8 infected cells, which is close to the molecular weight calculated for NS4, from the sequence of Seg-9 (,10 kDa). The anti-BTV NS4 antiserum is therefore specific to NS4 and does not cross react with other viral proteins. Western blot analysis using non-infected BHK-21 cells, showed that anti-BTV NS4 rabbit antiserum does not cross react with cellular proteins. A similar analysis, using antisera raised against recombinant GIV NS4, identified a protein of approximately 20 kDa in GIV infected cells (figure 5), corresponding to the theoretical molecular weight of NS4 deduced from the sequence of GIV Seg-9. The anti-GIV NS4 antiserum is therefore specific to GIV NS4 and does not cross react with other viral proteins. Western blot analysis using non-infected BHK-21 cells, showed that anti-GIV NS4 rabbit antiserum does not cross react with cellular proteins. Western blot analyses of purified BTV virus particles showed no reaction with anti-BTV NS4 antibodies, indicating that NS4 is 'non-structural' ( figure 6A, 6B ). Figure 7 shows infected and non-infected BHK-21 cells probed with anti-BTV-8 VP2 antibodies raised in mice. Figure 8 shows infected and non-infected BHK-21 cells probed with anti-BTV8 immune serum from experimentally infected mice. NS4 was identified in the nuclear fraction of BTV-infected BHK-21 cells harvested at 24 hours post-infection by western blot. Rabbit anti-BTV NS4 immune serum, identified the same band in the nuclear extract that was previously identified in infected cell lysates (figure 9). No band was identified in non-infected nuclear extracts. NS4 was detected as early as 4 hours post-infection, mainly in the cytoplasm of BHK-21 cells infected with BTV-8 or GIV ( figure 10A, 10B) . At 24 hours post-infection, NS4 formed small aggregates throughout the cytoplasm and nucleus, suggesting that it makes specific interactions with itself and/or other infected cell components ( figure 11A, 11B) . This is consistent with bioinformatic analyses which identified nuclear localisation signals in NS4 of GIV and BTV (as well as YUOV, PHSV, EHDV, AHSV). Although not all cells are morphologically intact at 72 hours postinfection, with GIV or BTV-8, at this stage NS4 was present in the cell membrane ( figure 12A, 12B) . This is consistent with bioinformatic analysis showing similarities between NS4 and membrane-associated proteins. Another set of cells, which were collected at 36 hours PI contained cells at different stages of infection. Those at an advanced stage of infection had depolymerised and depleted tubulin (figure 13A). No immuno-fluorescence signal was detected when non-infected cells were labelled using anti-NS4 antibodies (figure 13B, indicating that the anti-NS4 antiserum does not cross react with cellular proteins. Further analyses with confocal microscopy identified nucleolar fluorescence using anti-NS4 antibodies in cells infected with either BTV or GIV. Localisation of the NS4 to the nucleoli was visible by confocal fluorescence as well as by overlaying the fluorescence signal onto cells imaged by differential interference contrast microscopy ( figure 14A, 14B) . Localisation of NS4 to the nucleoli was confirmed using anti-fibrillarin antibodies, giving a fluorescence signal that was super-imposable on that of NS4 in the nucleoli ( figure 15 ). BHK-21 cells transfected with pCI-BTVNS4 or pCI-GIVNS4 resulted in expression of NS4 in both the cytoplasm and nucleus ( figure 16A, 16B) . NS4 was also detected in the nucleoli (figure 16B). Expressed NS4 was abundant in the cytoplasm where it formed aggregates similar to those found in infected cells. In many cells NS4 formed spherical bodies with 0.7 and 1 mm in diameter ( figure 16A and 16B) . Similar spherical bodies were occasionally also observed in cells infected with BTV-8 or GIV (figure 17A). Staining with the lipid stain oil-red-O, showed that these spherical bodies are associations between NS4 and lipid droplets ( figure 17B and figure 18A, 18B) . These bodies were identified in BTV-8 infected cells ( figure 17B and figure 18B) , where the oil-red-O stains lipids in the centre of the droplet while NS4 surrounds the lipid droplet. Figure 18A shows cells transfected with pCI-GIVNS4 stained with oil-red-O. Figure 19 shows non-infected cells stained with oil-red-O, where lipid droplets stain with red only. Similar data were recently reported for rotaviruses, where VP2, VP6 or NSP5 were found to associate with lipid droplets [39] . The mice immune serum from an animal infected with BTV-8 identified a protein in cells transfected with pCI-BTVNS4 expressing BTV-8 NS4. The protein band had the same size as that identified by the anti-BTV NS4 rabbit immune serum in BTV-infected cells. No band was identified in non-transfected cells (figure 20). Incubation of a dsRNA ladder (500-21 bp) with Dicer led to cleavage of long dsRNAs, generating 21 bp-long RNAs. Incubation of the dsRNA ladder with BTV NS4 or GIV NS4 alone did not alter dsRNA integrity. dsRNA preincubated with NS4 of GIV was protected against Dicer cleavage, consistent with previous findings regarding the presence of a dsRNA-binding domain. However, BTV NS4 did not protect dsRNA against Dicer and dsRNA was still processed into 21 bp long fragments, as analysed by agarose gel electrophoresis ( figure 21) . Incubation of dsRNA with BAV outer capsid protein VP9 (as a control) did not affect Incubation of a dsDNA ladder (2645-36 bp) with DNAse I led to degradation, while incubation with BTV or GIV NS4 only did not affect dsDNA integrity. However, dsDNA pre-incubated with either BTV or GIV NS4 was at least partially protected against In the colorimetric assay to detect NS4-dsRNA binding, the wells devoid of dsRNA were all negative, with a very low background (values close to zero). Negative control wells, containing only dsRNA, also had OD values close to zero (figure 23). Wells containing biotinilated dsRNA and BTV NS4 were also negative, indicating that BTV NS4 does not bind dsRNA. However, the wells containing dsRNA and GIV NS4, had increasing OD values with an almost linear relationship between the fixed NS4 concentration (150 ng/well) and the increasing dsRNA concentration, reaching a plateau at 320 ng of dsRNA/well (figure 23). This confirms the existence of a dsRNAbinding domain in GIV NS4, which is absent from BTV NS4. Within the family Reoviridae, genome segments encoding more than one protein, from distinct, ORFs have been previously reported for the aquareoviruses, fijiviruses, orthoreoviruses, rotaviruses, phytoreoviruses and oryzaviruses [1] . Genome segments of the orthoreoviruses, phytoreoviruses, oryzaviruses and rotaviruses can be bi-or tri-cistronic with overlapping ORFs. Those in the phytoreoviruses (Seg-9 and Seg-12), orthoreoviruses (segment S1) and rotaviruses (Seg-11) were also found to be expressed in infected cell cultures [40, 41, 42, 43, 44] . Translation of overlapping ORFs from reovirus genome segments has usually been shown to be dependent on leaky scanning [41, 43, 44] although scanning-independent ribosome shunting has also been described [42, 45] . An overlapping ORF in Seg-9, designated as ORFx, was recently identified by bioinformatic analysis in both insect-borne and tick-borne orbiviruses [17, 18] . ORFx appeared to encode a protein with the potential to bind dsRNA that was tentatively named as VP6db [18] . However, in line with previous orbivirus protein nomenclature, we have renamed these proteins, based on their theoretical size and absence from the virion, as non-structural protein 4 (NS4). GIV NS4 was previously shown to contain significant aa sequence matches with dsRNA-binding domains [18] . Further analyses of their amino acid sequences indicate that NS4 may be structured as 'coiled-coils' and that BTV NS4 exhibits significant relatedness (as identified by the pfam programme) with nucleic acid binding proteins that also have coiled-coils or helical structures and are associated with ER or cell membranes. It was suggested that translation of ORFx may be initiated via 'leaky ribosome-scanning' [17] , although the presence of additional AUG codons between the VP6 initiation codon and the presumed NS4 initiation codon in some orbiviruses (including BTV) suggests that additional mechanisms for bypassing intervening AUG codons may be operating [17] . An A-rich polypurine tract is present upstream of the NS4 ORF in all of the sequenced Orbivirus species, except the tick borne St. Croix River virus (SCRV, [46] . The SCRV NS4-ORF (nt 101-379) is interrupted by a stop codon at position 217. Although hydrophobicity profiles of putative NS4 proteins from BTV, AHSV, PHSV, YUOV and GIV as analysed using the Kyte and Doolittle algorithm [22] are somewhat variable, overall they show broadly similar patterns of conserved domains, indicating that the NS4 proteins are generally hydrophilic ( figure 24 ). BTV-8 infected KC or BHK-21 cells and GIV infected BHK-21 cells all contain NS4, as revealed by western blot analyses and confocal microscopy, confirming the existence of a new and previously un-described protein, encoded by ORFx of orbivirus Seg-9. NS4 has not previously been detected in purified BTV virus or core particles [20] . Western blot analyses of purified BTV-8 particles confirmed that the protein is 'non-structural'. Bioinformatic analyses indicate that NS4 contains coiled-coils and is structurally related to other mammalian proteins, with helical or coiled-coil regions. These analyses also suggest that the NS4 may be functionally related to proteins involved in nucleic acid binding, or associated with lipids and membranes. Nuclear localisation signals were predicted in NS4 of PHSV, YUOV, EHDV, BTV, AHSV and GIV. All these proteins are rich in arginine and lysine residues that are essential for NLS [47] . Double-stranded RNA-binding proteins (DRBPs) do not recognize specific nucleotide sequences but interact primarily with A-form double helix RNAs, which differ from the typical dsDNA B-form helix in that the minor groove is shallow and broad while the major groove is narrow and deep. This conformation allows DRBPs to bind non-specifically to dsRNAs. Indeed, the lack of nucleotide binding recognition suggests that target specificity may generally be governed through interactions with other proteins, since many DRBPs bind strongly but nonspecifically to any dsRNA structure in vitro. GIV NS4 can protect dsRNA from degradation by RNAse III endoribonucleases, confirming previous sequence analyses indicating the presence of a dsRNA-binding domains [18] . BTV NS4 which is half the theoretical size of its counterpart in GIV, lacks dsRNA binding domains and did not protect dsRNA from Dicer or RNAse III. NS4 of GIV and BTV both failed to protect ssRNA or ssDNA from degradation by RNAse A, or nuclease S1 respectively (data not shown). However, NS4 of both GIV and BTV did protect dsDNA from degradation by DNAse I, indicating an ability to bind dsDNA. Fluorescent confocal microscopy confirmed that NS4 is expressed in both BTV and GIV infected cells, and starts to accumulate in the cytoplasm and nucleus (as fine aggregates) as early as 4 hours post-infection. However, at 72 hours postinfection NS4 was associated with the cell membrane. This is consistent with analyses suggesting similarities between NS4 and ER-lipid-or membrane-associated proteins [37] . Cells infected with BTV or GIV or transfected with plasmids expressing NS4, showed interaction of NS4 with lipid droplets within the cytoplasm. This is consistent with bioinformatic analysis that identified similarities between NS4 and lipid-associated domains. Similar data were recently reported for VP2, VP6 and NSP5 of rotavirus [39] . . Colorimetric assay to detect interactions of NS4 with dsRNA. The graph shows colorimetric OD readings plotted against concentrations of dsRNA. Increasing concentrations (from 1 to 640 ng) of a biotinylated dsRNA were bound to wells coated with streptavidin. BTV NS4 or GIV NS4 were added to the wells in triplicate. Wells not containing dsRNA/NS4 were included as negative controls. Wells from which dsRNA was omitted, but in which NS4 (BTV or GIV) alone was incubated were also included as controls. Only wells containing the dsRNA to which GIV NS4 was added reacted with anti-GIV antibodies as indicated by increasing OD readings. The readings were almost linear (reaching a plateau at 320 ng of dsRNA) indicating that dsRNA acted as a target for binding of GIV NS4. doi:10.1371/journal.pone.0025697.g023 Viruses can interact with components of the nucleolus [48, 49] and viral proteins can co-localise with proteins such as nucleolin, B23 and fibrillarin (components of the nucleolus). The use of antifibrillarin antibodies identified NS4 in the nucleoli of cells harvested at 24 hours post-infection. Although NS4 was detected in the nucleoli late in infection, anti-fibrillarin antibodies failed to detect fibrillarin. This may reflect BTV induced apoptosis, leading to nuclear condensation and DNA fragmentation, blebbing of the plasma membrane and shrinkage [50, 51] , and/or host cell shut-off [4] . Similar findings were reported in rotavirus (another member of the family Reoviridae, genus Rotavirus) where NSP2 protein was found to cause depolymerisation of tubulin [52] . As part of their replication strategy, viruses can use nucleolar components to favour viral transcription and translation, or alter the cell cycle [48, 49] . Western blot analysis indentified NS4 in the nuclear fraction of BTV infected cells, while immunofluorescence confocal microscopy co-localised NS4 to the nucleolus. GIV-NS4 showed sequence similarity to UTP20, a small subunit processome component and a component of the nucleolus. UTP20 is part of the U3 small nucleolar RNA (snoRNA) protein complex (U3 snoRNP) and is involved in 18S rRNA processing [53] . Whether NS4 interferes with the processing of the 18s rRNA remains to be clarified in future work. The ability of GIV NS4 to protect dsRNA from cleavage by endoribonucleases of the RNAse III family and its ability to bind dsRNA agree with sequence analyses that indicated the presence of a dsRNA binding domain in GIV NS4 [18] . The inability of BTV NS4 to protect dsRNA from cleavage by endoribonucleases of the RNAse III family and its inability to bind dsRNA in a platebased colorimetric assay are in agreement with sequence analyses that failed to detect a dsRNA binding domain in its aa sequence [18] . Other reoviruses dsRNA-binding proteins include Sigma 3 of mammalian orthoreovirus (found in both the cytoplasm and nucleus), and pns10 of rice dwarf virus [54, 55] . SCRV, which persistently infects tick cells but does not grow in mammalian cells, appears to have a non-functional NS4 ORF that is interrupted by a stop codon. These observations suggest that NS4 expression could play a role in productive infection of mammalian cells. The data presented here show that the orbivirus genome encodes four distinct non-structural proteins (NS1-NS4). NS1 and NS3 play an important role in orbivirus exit mechanisms from infected cells [15] . BTV infects mammalian cells, usually resulting in a lytic infection, while infection of KC cells derived from the BTV vector Culicoides sonorensis, become persistently infected with little or no evidence of cell lysis [56, 57] . Previous work showed that intracellular expression of an NS1 specific antibody fragment (scFv) destabilised the formation of NS1 tubules in BTV infected cells [14] . As a consequence, cells became persistently infected and viruses exited by budding instead of via cell lysis. Although BTV NS3 is effectively expressed in insect cells [15] , it is much less abundant in mammalian cells [4] . It was suggested previously [14] that the relative levels of NS1 to NS3 synthesised during infection dictate the fate of cellular pathogenesis as of whether the virus exit occurs by lysis or budding. The rapid accumulation of NS4 in the cytoplasm as early as 4 hours post-infection suggests that this protein plays an early role in the virus replication cycle. At 72 hours post-infection NS4 was absent from the nucleus which could be the consequence of changes affecting the nucleus and the integrity of the nuclear membrane. The presence of the NS4 in the plasma membrane late in infection suggests that it may play a role, alongside NS1 and NS3, in virus exit. Further co-localisation studies will be carried out to assess NS4 interactions with other viral or cellular protein components.

Geographic Distribution and Risk Factors of the Initial Adult Hospitalized Cases of 2009 Pandemic Influenza A (H1N1) Virus Infection in Mainland China

BACKGROUND: As of 31(st) March 2010, more than 127,000 confirmed cases of 2009 pandemic influenza A (H1N1), including 800 deaths, were reported in mainland China. The distribution and characteristics of the confirmed cases in the initial phase of this pandemic in this country are largely unknown. The present study aimed to characterize the geographic distribution and patient characteristics of H1N1 infection in the 2009 pandemic as well as to identify potential risk factors associated with adverse patient outcome in China, through retrospective analyses of 885 hospitalized cases with confirmed H1N1 infection. METHODOLOGY/PRINCIPAL FINDINGS: The proportional hazards model was employed to detect risk factors for adverse outcome; the geo-statistical maps were used to characterize the distribution of all 2668 confirmed H1N1 patients throughout mainland China. The number of new cases increased slowly in May, 2009, but rapidly between June and August of the year. Confirmed cases were reported in 26 provinces; Beijing, Guangdong, Shanghai, Zhejiang and Fujian were the top five regions of the incidence of the virus infection. After being adjusted for gender, age, chronic pulmonary disease and other general symptoms, delay for more than two days before hospital admission (HR: 0.6; 95%CI: 0.5–0.7) and delayed onset of the H1N1-specific respiratory symptoms (HR: 0.3; 95%CI: 0.2–0.4) were associated with adverse patient outcome. CONCLUSIONS/SIGNIFICANCE: The 2009 pandemic influenza A affected east and southeast coastal provinces and most populous cities more severely than other regions in mainland China due to higher risk of high level traffic-, high population density-, and high population mobility-associated H1N1 transmission.The clinical symptoms were mild in the initial phase of infection. Delayed hospital admission and delayed appearance of respiratory symptoms were among the major risk factors for poor patient outcome. These findings may have significant implications in the future pandemic preparedness and response.

The 2009 pandemic influenza A (H1N1) was first reported in Mexico and then rapidly spread around the world, unfortunately due to the 'convenience' of airplane-based modern transportation system. On 11 th June 2009, the World Health Organization (WHO) raised the warning level to the sixth phase, indicating that the episode of influenza had entered a pandemic stage [1] . As of 11 th April 2010, more than 214 countries and territories or communities had reported laboratory-confirmed cases of 2009 H1N1, including over 17798 deaths [2] . The first case in mainland china was reported on May 11 th , 2009. The Ministry of Health of the People's Republic of China released Guidelines for H1N1 Influenza Diagnosis and Treatment in the middle of July, 2009 [3] . By March 31 st 2010, more than 127,000 laboratorial confirmed cases including 800 deaths were reported in the country [4] . Since the H1N1 pandemic in 2009, numerous field investigations and epidemiological studies have investigated the spatial-temporal dynamics, geographic distribution and patient characteristics of the 2009 H1N1 pandemic. Through analyses of a total of 377 H1N1associated deaths in the United States, Fowlkes et al. demonstrated that the H1N1-associated mortality rate varied substantially in different geographic regions (i.e. highest in Hawaii, New York and Utah) and in different age groups of the infected population (i.e. 76% in patients aged 18-65 years and 9% in patients aged $ 65 years) [5] . Similar geographic region-and patient age-dependent incidence of H1N1 infection and subsequent mortality have also been demonstrated in South America [6] , and Australia [7] . In addition, the susceptibility to H1N1 infection may vary in different races; Wenger et al. shown that Alaska Native people and Asian/ Pacific Islanders (A/PI) were 2-4 times more likely to be infected by H1N1 virus and hospitalized than white Caucasians [8] . To date, little is known regarding the spatio-temporal characteristics of the 2009 pandemic H1N1 as well as factors affecting the recovery of the infected patients in mainland China, especially for the initial cases, who were younger and more mobile than general population. The present study was therefore conducted to investigate the geographic distribution of H1N1 infection in the 2009 pandemic as well as the risk factors for adverse patient outcome after initial H1N1 virus infection in mainland China. Our goal was to generate solid scientific bases for adequate preparedness for potential H1N1 pandemic in the future. Basic characteristics/denigraphics of the patients Given the unique nature of 2009 pandemic H1N1, both suspected and laboratory confirmed cases had to be reported to the Chinese Centre for Disease Control and Prevention (CCDC). During May 7 th and August 12 th , 2009, the number of reported laboratory-confirmed H1N1 cases was 2668. In this study, we analyzed the data of 885 adult hospitalized patients, accounting for 68.6% of the total confirmed cases. The major demographic and clinical characteristics of these patients are summarized in Table 1 . Of the 885 hospitalized adult cases, 590 (66.7%) experienced a delay#2 days in hospital admission, 815 (92.1%) completely recovered and were free of H1N1-associated symptoms on discharge whereas 70 (7.9%) were censored (not recovered when left hospitals). The median time from illness onset to hosptial discharge was 7 days (inter-quartile range 2-22). The analyses using the proportional hazards model showed that gender, time from illness onset to hospital admission, general symptoms and respiratory symptoms were all related to the outcome of the H1N1 patients (Table 2) . Compared with male patients, females patients had a lower porpotion of recovery (HR 0.9, 95%CI: 0.7-1.0), but the differnence was not statistically significant. Delay in hospital admission (longer than 2 days) decreased the probability of recovery (HR 0.6, 95%CI: 0.5-0.7). General symptoms and respiratory symptoms were adversely associated with the rate of recovery in hospital with HR of 0.7 (95%CI: 0.5-1.0) and 0.3 (95%CI: 0.2-0.4), respectively. Showed in Figure 1 are the cumulative resolution rates. It is apparent that the cumulative resolution rate was lower in patients with delayed hospital admission than in those promptly admitted (x 2 = 60.978, P,0.001), and lower in patients with general symptoms (x 2 = 3.977, P = 0.046) or respiratory symptoms (x 2 = 100.261, P,0.001) than in those without corresponding symptoms. No significant difference was observed between genders (x 2 = 1.634, P = 0.201). The weekly reported new cases of confirmed H1N1 infection in different geographic regions of mainland China during May 7 th to August 12 th , 2009 are presented in Figure 2 . A total of 26 provinces or cities had confirmed cases of infection, among which Beijing city had the highest number of infection (670 cases), followed by Guangdong province(628 cases), Shanghai city(318 cases), Zhejiang province (214 cases), and Fujian province(195 cases). The number of new cases increased slowly in May 2009, but rapidly between June and August. In May, 87.3% of the cases were imported ones, while the domestic cases (67.3%) became dominant between June and August. Most of the provinces had reported laboratory-confirmed cases by August. However, the cases were more likely to occur in the southeast coastal areas or areas that serve as major transport stations with relatively prosperous economy, convenient transport system and greater population. In this study, patients with leukocyte and neutrophils abnormalities as well as patients undergoing antivirus treatment were excluded from the Cox regression model. In seasonal influenza, cough, mild leukopenia, and mild C-reactive protein elevation are relatively common clinical manifestations [9] . Pathologically, the H1N1 virus has features similar to those encountered in the infection with other highly virulent influenza A viruses such as the 1918 H1N1 and H5N1 viruses [10] . However, as the disease was generally a mild procedure, typical inflammation molecules might not be qualified to serve as diagnostic markers. For patients with complications, the secondary bacterial infection which causes elevation in leucopenia and C-reactive protein is most likely to occur. The widely used antivirus treatment like oseltamivir might increase the proportion of oseltamivir-resistant influenza A. Therefore, alternative treatment and vaccination (although, low coverage) are highly recommended. It has been documented that the adverse outcome, particularly a severe form (i.e. entering ICU or death), after influenza infection is associated with numerous risk factors, including age and underlying medical conditions of the patients. Nolan et al. in a recent study on pandemic or seasonal influenza have observed that infants, the elderly and people suffering from chronic diseases are of high risks for adverse outcome [11] ; some other authors have demonstrated significant adverse clinical outcome in patients aged $20 years and patients delayed in hospital admission [12, 13] . Nevertheless, only adult hospitalized cases were included in the present study. With the fast-growing public transportation and increasing socio-economic activities, population mobility has become a concerned problem in the prevention of infectious diseases. Ever since the outbreak of SARS, the central role that Geographic Information System (GIS) plays in the early detection and rapid response to infectious diseases has become evident to researchers as well as policy markers [14, 15] . However, how to integrate GIS into the traditional epidemiological model is a key challenge [16] . Use of GIS in our analyses in this study showed that the incidence of H1N1 infection in the 2009 pandemic in mainland China was higher in the coastal provinces in the southeast as well as in Beijing and Shanghai, two of the most populous cities in the country, where the infection occurred in clusters in the initial phase, followed by spread to adjacent provinces. Moreover, the analyses also demonstrated that within some of the provinces with confirmed cases, the number of H1N1 infection varied widely as well. In most cities and districts, there was still no H1N1 confirmed cases, the overall spread level of H1N1 virus was still regional. Nevertheless, our analyses on the H1N1 spread and the geographic distribution were qualitative only; further quantitative analyses such as correlation or Poisson regression are needed to test the hypothesis that the H1N1 virus spreads more quickly in the crowd area with more convenient travel system. The limitation of this study is that it only enrolled the adult hospitalized patients. This study included no pediatric cases and therefore was unable to assess the differences in risk factors for adverse patient outcome between children and adults. In the initial phase of H1N1 infection in mainland China, hospital admission was required of every confirmed case. The information on the confirmed cases was from the Chinese Infectious Disease Surveillance System. Sub-clinical infection was common in H1N1 infection. People with mild symptoms might not be included in the Chinese Infectious Disease Surveillance System, possibly causing bias in our analyses. Further studies were needed in the mathematical model and simulation for the transmission characteristics. Despite the limitations, this study through retrospectively analyses characterized clinical features, disease course, and hospitalization period of the 2009 pandemic H1N1 in mainland China. In adult H1N1 patients, age, chronic pulmonary disease and the appearance of general symptoms, delayed hospital admission and the appearance of respiratory symptoms but not gender were independent risk factors for adverse outcome. Our study also documented the spatial and temporal characteristics of the 2009 pandemic H1N1 at its early stage in mainland China. The study was approved by ethic committees of both Capital Medical University in Beijing, China and the Chinese Centre for Disease Control and Prevention. Data on a total of 885 adult confirmed H1N1 cases from the CCDC. A confirmed case was defined as a person with influenzalike clinical manifestations and positive laboratory test for H1N1 virus, as assessed with the use of a reverse-transcriptasepolymerase-chain-reaction (RT-PCR) assay. Those patients who were still asymptomatic but positive for the RT-PCR assay were quarantined. The cases included in this study were those reported in the early phase of the epidemic between May 7 th and August 16 th , 2009 to the CCDC by sub-CCDC offices after face to face interviews with the individuals of suspected infection. The case report form was used, which included the following information: gender, age, chronic pulmonary disease history, time from onset to hospital admission, general symptoms (at least one of the following: body temperature$37.5uC, dizziness, headache, chill, myalgia, arthralgia, weakness, anorexia, and/or conjunctivitis), respiratory symptoms (at least one of the following: runny nose, nasal obstruction, sneezing, dry cough, sputum, sore throat, pharyng itching, shortness of breath, and/or chest pain), pharyngeal abnormality, tonsil abnormality, white blood cells count, complication(s) (at least one of pneumonia, liver function impairment, and/or myocardic injury), antiviral medication and outcome. Recovery was defined as resolution of symptoms as well as negative test results for H1N1 virus RNA for two consecutive days in throat swabs. Survival analysis was used to evaluate adverse factors for recovery (days from illness onset to recovery) in the early phase of H1N1 epidemic. Potential risk factors assessed included gender, age, chronic pulmonary disease, time from onset to hospital admission, general symptoms, respiratory symptoms, pharyngeal abnormality, tonsil abnormality, white blood cells count, complication and antiviral medication. The parameters of the recovery model were estimated by backward stepwise likelihood ratio test in proportional hazards model. Censored patient was considered as not recovered. Twosided P values and 95% confidence interval of the parameters were calculated. The cumulative resolution rate was analyzed using Kaplan-Meier analysis and log-rank test and compared between subgroups of patients. The subgroups were based on gender, time from onset to hospital admission, general symptoms or respiratory symptoms. Geo-statistical maps were used to characterized the spatial and temporal distribution of weekly confirmed cases during May 7 th and August 12 th . Longitude and latitude coordinates of each province were identified from a national GIS database. The procedure was fulfilled in a GIS (ArcInfo 9.2, ESRI, Redlands, CA) by overlaying a national vector map. All statistical analyses were conducted with the SPSS 13.0 software for Windows (SPSS Inc, Chicago, IL, USA).

Inferring viral quasispecies spectra from 454 pyrosequencing reads

BACKGROUND: RNA viruses infecting a host usually exist as a set of closely related sequences, referred to as quasispecies. The genomic diversity of viral quasispecies is a subject of great interest, particularly for chronic infections, since it can lead to resistance to existing therapies. High-throughput sequencing is a promising approach to characterizing viral diversity, but unfortunately standard assembly software was originally designed for single genome assembly and cannot be used to simultaneously assemble and estimate the abundance of multiple closely related quasispecies sequences. RESULTS: In this paper, we introduce a new Viral Spectrum Assembler (ViSpA) method for quasispecies spectrum reconstruction and compare it with the state-of-the-art ShoRAH tool on both simulated and real 454 pyrosequencing shotgun reads from HCV and HIV quasispecies. Experimental results show that ViSpA outperforms ShoRAH on simulated error-free reads, correctly assembling 10 out of 10 quasispecies and 29 sequences out of 40 quasispecies. While ShoRAH has a significant advantage over ViSpA on reads simulated with sequencing errors due to its advanced error correction algorithm, ViSpA is better at assembling the simulated reads after they have been corrected by ShoRAH. ViSpA also outperforms ShoRAH on real 454 reads. Indeed, 7 most frequent sequences reconstructed by ViSpA from a real HCV dataset are viable (do not contain internal stop codons), and the most frequent sequence was within 1% of the actual open reading frame obtained by cloning and Sanger sequencing. In contrast, only one of the sequences reconstructed by ShoRAH is viable. On a real HIV dataset, ShoRAH correctly inferred only 2 quasispecies sequences with at most 4 mismatches whereas ViSpA correctly reconstructed 5 quasispecies with at most 2 mismatches, and 2 out of 5 sequences were inferred without any mismatches. ViSpA source code is available at http://alla.cs.gsu.edu/~software/VISPA/vispa.html. CONCLUSIONS: ViSpA enables accurate viral quasispecies spectrum reconstruction from 454 pyrosequencing reads. We are currently exploring extensions applicable to the analysis of high-throughput sequencing data from bacterial metagenomic samples and ecological samples of eukaryote populations.

Results: In this paper, we introduce a new Viral Spectrum Assembler (ViSpA) method for quasispecies spectrum reconstruction and compare it with the state-of-the-art ShoRAH tool on both simulated and real 454 pyrosequencing shotgun reads from HCV and HIV quasispecies. Experimental results show that ViSpA outperforms ShoRAH on simulated error-free reads, correctly assembling 10 out of 10 quasispecies and 29 sequences out of 40 quasispecies. While ShoRAH has a significant advantage over ViSpA on reads simulated with sequencing errors due to its advanced error correction algorithm, ViSpA is better at assembling the simulated reads after they have been corrected by ShoRAH. ViSpA also outperforms ShoRAH on real 454 reads. Indeed, 7 most frequent sequences reconstructed by ViSpA from a real HCV dataset are viable (do not contain internal stop codons), and the most frequent sequence was within 1% of the actual open reading frame obtained by cloning and Sanger sequencing. In contrast, only one of the sequences reconstructed by ShoRAH is viable. On a real HIV dataset, ShoRAH correctly inferred only 2 quasispecies sequences with at most 4 mismatches whereas ViSpA correctly reconstructed 5 quasispecies with at most 2 mismatches, and 2 out of 5 sequences were inferred without any mismatches. ViSpA source code is available at http://alla.cs.gsu.edu/~software/VISPA/vispa.html. Conclusions: ViSpA enables accurate viral quasispecies spectrum reconstruction from 454 pyrosequencing reads. We are currently exploring extensions applicable to the analysis of high-throughput sequencing data from bacterial metagenomic samples and ecological samples of eukaryote populations. Many viruses (including SARS, influenza, HBV, HCV, and HIV) encode their genome in RNA rather than DNA. Unlike DNA viruses, RNA viruses lack the ability to detect and repair mistakes during replication [1] and, as a result, their mutation rate can be as high as 1 mutation per each 1,000-100,000 bases copied per replication cycle [2] . Many of the mutations are well tolerated and passed down to descendants, producing a family of co-existing related variants of the original viral genome referred to as quasispecies, a concept that originally described a mutation-selection balance [3] [4] [5] [6] [7] . The diversity of viral sequences in an infected individual can cause the failure of vaccines and virus resistance to existing drug therapies [8] . Therefore, there is a great interest in reconstructing genomic diversity of viral quasispecies. Knowing sequences of the most virulent variants can help to design effective drugs [9, 10] and vaccines [11, 12] targeting particular viral variants in vivo. Briefly, the 454 pyrosequencing system shears the source genetic material into fragments of approximately 300-800 bases. Millions of single-stranded fragments are sequenced by synthesizing their complementary strands. Repeatedly, nucleotide reagents are flown over the fragments, one nucleotide (A, C, T, or G) at a time. Light is emitted at a fragment location when the flown nucleotide base complements the first unpaired base of the fragment [13, 14] . Multiple identical nucleotides may be incorporated in a single cycle, in which case the light intensity corresponds to the number of incorporated bases. However, since the number of incorporated bases (referred to as a homopolymer length) cannot be estimated accurately for long homopolymers, it results in a relatively high percentage of insertion and deletion sequencing errors (which respectively represent 65%-75% and 20%-30% of all sequencing errors [15, 16] ). The software provided by instrument manufacturers were originally designed to assemble all reads into a single genome sequence, and cannot be used for reconstructing quasispecies sequences. Thus, in this paper we address the following problem: Given a collection of 454 pyrosequencing reads generated from a viral sample, reconstruct the quasispecies spectrum, i.e., the set of sequences and the relative frequency of each sequence in the sample population. A major challenge in solving the QSR problem is that the quasispecies sequences are only slightly different from each other. The amount and distribution along the genome of differences between quasispecies varies significantly between virus species, as different species have different mutation rates and genomic architectures. In particular, due to the lower mutation rate and longer conserved regions, HCV quasispecies are harder to reconstruct than quasispecies of HBV and HIV. Additionally, the QSR problem is made difficult by the limited read length and relatively high error rate of high throughput sequencing data generated by current technologies. The QSR problem is related to several well-studied problems: de novo genome assembly [17] [18] [19] , haplotype assembly [20, 21] , population phasing [22] and metagenomics [23] . As noted above, de novo assembly methods are designed to reconstruct a single genome sequence, and are not well-suited for reconstructing a large number of closely related quasispecies sequences. Haplotype assembly does seek to reconstruct two closely related haplotype sequences, but existing methods do not easily extend to the reconstruction of a large (and a priori unknown) number of sequences. Computational methods developed for population phasing deal with large numbers of haplotypes, but rely on the availability of genotype data that conflates information about pairs of haplotypes. Metagenomic samples do consist of sequencing reads generated from the genomes of a large number of species. However, differences between the genomes of these species are considerably larger than those between viral quasispecies. Furthermore, existing tools for metagenomic data analysis focus on species identification, as reconstruction of complete genomic sequences would require much higher sequencing depth than that typically provided by current metagenomic datasets. In contrast, achieving high sequencing depth for viral samples is very inexpensive, owing to the short length of viral genomes. Mapping based approaches to QSR are naturally preferred to de novo assembly since reference genomes are available (or easy to obtain) for viruses of interest, and viral genomes do not contain repeats. Thus, it is not surprising that such approaches were adopted in the two pioneering works on the QSR problem [24, 25] . Eriksson et al. [24] proposed a multi-step approach consisting of sequencing error correction via clustering, haplotype reconstruction via chain decomposition, and haplotype frequency estimation via expectation-maximization, with validation on HIV data. In Westbrooks et al. [25] , the focus is on haplotype reconstruction via transitive reduction, overlap probability estimation and network flows, with application to simulated error-free HCV data. Recently, the QSR software tool ShoRAH was developed [26] and applied to HIV data [27] . Another combinatorial method for QSR was also developed and applied to HIV and HBV data in [28] , with results similar to those of ShoRAH. Our contributions in this paper are as follows: • A novel QSR tool called Viral Spectrum Assembler (ViSpA) taking into account sequencing errors at multiple steps, • Comparison of ViSpA with ShoRAH on HCV synthetic data both with and without sequencing errors, and • Statistical and experimental validation of the two methods on real 454 pyrosequencing reads from HCV and HIV samples. Our method for inferring the quasispecies spectrum of a virus sample from 454 pyrosequencing reads consists of the following steps (see Fig. 1 ): • Constructing the consensus virus genome sequence for the given sample and aligning the reads onto this consensus, • Preprocessing aligned reads to correct sequencing errors, • Constructing a transitively reduced read graph with vertices representing reads and edges representing overlaps between them, • Selecting paths in the read graph that correspond to the most probable quasispecies sequences, and assembling candidate sequences for selected paths by weighted consensus of reads, and • Estimating candidate sequence frequencies by EM Below we describe each step separately. We assume that a reference genome sequence of the particular virus strain is available (e.g., from NCBI [29] ). Since viral genomes do not have sizable repeats and the quasispecies sequences are usually close enough to the reference sequence, the majority of reads can typically be uniquely aligned onto the reference genome. However, a significant number of reads may remain unaligned due to differences between the reference genome and sequences in the viral sample. In order to recover as many of these reads as possible, we iteratively construct a consensus genome sequence from aligned reads. In particular, we first align 454 pyrosequencing reads to the reference sequence using the SEGEMEHL software [30] . Then we extend the reference sequence with a placeholder I for each nucleotide inserted by at least one uniquely aligned read. Similarly, we add a placeholder D to the read sequence for each reference nucleotide missing from the aligned read. Then we perform sequential multiple alignment of the previously aligned reads against this extended reference sequence. Finally, the consensus genome sequence is obtained by (1) replacing each nucleotide in the extended reference with the nucleotide or placeholder in the majority of the aligned reads and (2) removing all I and D placeholders, respectively corresponding to rare insertions and to deletions found in a majority of reads. Reads may contain a small portion of unidentified nucleotides denoted by N'swe treat N as a special allele value matching any of nucleotides A, C, T, G, as well as placeholders I, and D. Iteratively, we replace the reference with the consensus and try to align the reads, for which we could not find any acceptable alignment previously. Our experiments on a dataset consisting of approximately 31,000 454 pyrosequencing reads generated from a 5.2kb-long HCV fragment (see data description in Results and Discussions) show that 85% of reads are uniquely aligned onto the reference sequence and an additional 9% of the reads are aligned onto the final consensus sequence. Reads that cannot be aligned onto the final consensus are removed from the further consideration. Since aligned reads contain insertions and deletions, we use placeholders I and D to simplify position referencing among the reads. All placeholders are treated as additional allele values but they are removed from the final assembled sequences. First, we substitute each deletion in the aligned reads with placeholder D. Deletion supported by a single read is replaced either with the allele value, which is present in all other reads overlapping this position, or with N, signifying an unknown value, otherwise. Next, we fill with placeholder I each gap in a read corresponding to the insertions in the other reads. All insertions supported by a single read are removed from consideration. We begin with the definition of the read graph, introduced in [25] and independently in [24] , and then describe the adjustments that need to be made to read graph construction and edge weights to account for sequencing errors as well as the high mutation rate between quasispecies. The read graph G = (V, E) is a directed graph with vertices corresponding to reads aligned with the consensus sequence. For a read u, we denote by b(u), respectively e(u), the genomic coordinate at which the first, respectively the last, base of u gets aligned. A directed edge (u, v) connects read u to read v if a suffix of u overlaps with a prefix of v and they coincide across the overlap. Two auxiliary vertices -a source s and a sink t are added such that s has edges into all reads with zero indegree and t has edges from all reads with zero outdegree. Then each st-path corresponds to a possible candidate quasispecies sequence. The read graph is transitively reduced, i.e., each edge e = (u, v) is removed if there is a uv-path not including edge e. Note that certain reads can be completely contained inside other reads. Let a superread refer to a read that is not contained in any other read and let the rest of the reads be called subreads. Subreads are not used in the construction of the read graph, but are taken into account in the final assembly of candidate sequences and frequency estimation. Since the number of different st-paths is exponential, we wish to generate a set of paths that have high probability to correspond to real quasispecies sequences. In order to estimate path probability, we independently estimate for each edge e the probability p(e) that it connects two reads from the same quasispecies, and then multiply estimated probabilities for all edges on the path. Under the assumption of independence between edges, if we assign to each edge e a cost equal tolog (p(e)) = log(1/p(e)), then the minimum-cost st-path will have the maximum probability to represent a quasispecies sequence. For reads without errors, [25] estimated the probability that two reads u and v connected by edge (u, v) belong to the same quasispecies as is the overhang between reads u and v [25] , N = #reads, q = #quasispecies, and L = #starting positions. Thus, in this case the cost of an edge with overhang Δ can be approximated by Δ ∝ log(1/p Δ ). To account for sequencing errors, we adjust the construction of the read graph to allow for mismatches. We use three parameters: (1) n = #mismatches allowed between a read and a superread, (2) m = #mismatches allowed in the overlap between two adjacent reads, and (3) t = #mismatches expected between a read and a random quasispecies. The probability that two reads u and v with j mismatches within an overlap of length o = e(u) b(v) belong to the same quasispecies can be estimated as: where ε is the estimated 454 sequencing error rate. As in the case of error-free reads, defining the edge costs as ensures that stpaths with low cost correspond to most likely quasispecies sequences. To generate a set of high-probability (low-cost) paths that are rich enough to explain observed reads, we compute for each vertex in the read graph the minimum cost st-path passing through it. Finding these paths is computationally fast. Indeed, we only need to compute two shortest-paths trees in G, one outgoing from s and one incoming into t; the shortest st-path passing through a vertex v is the concatenation of the shortest s v-and vt-paths. Preliminary simulation experiments (see Additional File 1) show that better candidate sets are generated when edge costs c defined by (1) and (2) are replaced by e c . In fact, if we use even faster dependency on c then we obtain better candidate sets. The fastest growing cost effectively changes the shortest path into so called maxbandwidth path, i.e., paths that minimizes maximum edge cost for the entire path and for each subpath. So, ViSpA generates candidate paths using this strategy. When no mismatches are allowed in the construction of the read graph, finding the candidate sequence corresponding to a st-path is trivial, since by definition adjacent superreads coincide across their overlap. When mismatches are allowed, we first assemble a consensus sequence from superreads used by the st-path. It may be not the best choice, especially when the coverage with superreads is low. Hence, we replace each initial candidate sequence with a weighted consensus sequence obtained using both superreads and subreads of the path, as described below. For each read r, we compute the probability that it belongs to a particular initial candidate sequence s as: where l and L denote the lengths of the read and initial candidate sequence, respectively, k is the number of mismatches between the read and the initial candidate sequence s, and t/L is the estimated mutation rate. Then final candidate sequence is computed as the weighted consensus over all reads, where the weight of a read is the probability that it belongs to the sequence. Note that, unlike the case without mismatches, the same candidate sequence can be obtained from different candidate st-paths, so we remove duplicates at the end of this step. We assume that reads R with observed frequencies where generated from a quasispecies population Q as follows. First, a quasispecies sequence q Q is randomly chosen accordingly to its unknown frequency f q . A read starting position is generated from the uniform distribution and then a read r is produced from quasispecies q with j sequencing errors. The probability of this event is calculated as h q r j l lj j , ( ) where l is the read length and ε is the sequencing error rate. In our simulation studies we use the following read data sets. In order to perform cross-validation on the assembly method, we simulate reads data from 1739-bp long fragment from the E1E2 region of 44 HCV sequences [32] when sequence frequencies are generated according to some specific distribution. In our simulation experiments, we use geometric distribution (i-th sequence is constant factor more frequent than the (i + 1)-th sequence) to create sample quasispecies populations with different number of randomly selected above-mentioned quasispecies sequences. We first simulate reads without sequencing errors: the length of a read follows normal distribution with a particular mean value and variance 400, and a starting position follows the uniform distribution. This simplified model of reads generation has two parameters: number of the reads that varies from 20K up to 100K and the average read length that varies from 200bp up to 600bp. Additionally, we simulate 454 pyrosequencing reads from 10 quasispecies sequences (following geometric distribution of frequencies) out of 44 HCV sequences [32] using FlowSim [33] . We generated 30K reads with average length 350bp. The data set Data1 has been received from HCV Research Group in Institute of Biomedical Research, at University of Birmingham. Data1 contains 30,927 reads obtained from the 5.2kb-long fragment of HCV-1a genome (which is more than a half of the entire HCV genome). The average (aligned) read length average is 292bp but it significantly varies as well as the depth of position coverage (see Additional File 1 for details). The depth of reads coverage variability is due to a strong bias in the sequence start points, reflecting the secondary structure of the template DNA or RNA used to generate the initial PCR products. As a result, shorter reads are produced by GC-rich sequences. Data1 is available upon request from the authors. The HIV dataset [27] contains 55,611 reads from mixture of 10 different 1.5kb-long region of HIV-1 quasispecies, including pol protease and part of the pol reverse transcriptase. The aligned reads length varies from 35bp to 584bp with average about 345bp (see Additional File 1 for details). In contrast to [27] , we do not filter out reads with low-quality scores. In all our experimental validations, we compare the proposed algorithm ViSpA with the state-of-the-art tool ShoRAH as well as with ViSpA on ShoRAH-corrected reads (ShoRAHreads + ViSpA). We say the quasispecies sequence is captured if one of the candidate sequences exactly matches it. We measure the quality of assembling by portion of the real quasispecies sequences being captured by candidate sequences (sensitivity = + ) and its portion among candi- Here, we see advantage of ViSpA over ShoRAH. Following [24] , we measure the prediction quality of frequency distribution with Kullback-Leibler divergence, or relative entropy. Given two probability distributions, relative entropy measures the "distance" between them, or, in the other words, the quality of approximation of one probability distribution by the other distribution. Formally, the relative entropy between true distribution P and approximation distribution Q is given by the formula: where summation is over all reconstructed original sequences I = {i | P(i) > 0, Q(i) > 0} , i. e., over all original sequences that have a match (exact or with at most k mismatches) among assembled sequences. The relative entropy is decreasing with increasing of the average read length. It is expected since sensitivity is increasing with increasing of the average read length and EM predicts underlying distribution more accurately. ViSpA algorithm considerably outperforms ShoRAH (see Fig. 2 (right)). However, ShoRAH has a significant advantage over ViSpA on a read data simulated by FlowSim both in prediction power and in robustness of results (see Table 1 ). Indeed, ShoRAH correctly infers 3 out of 10 real quasispecies sequences whereas ViSpA reconstructs only 1 sequence. Additionally, 10 most frequent assemblies inferred by ShoRAH are more robust with repeating up to 45% of times on 10%-reduced data versus 1% of times for ViSpA's assemblies. This advantage can be explained by superior read correction in ShoRAH. If ViSPA is used on ShoRAH-corrected reads, the results drastically improves: 5 quasispecies sequences are inferred and exactly 95% of times are repeated on reduced data, confirming that ViSpA is better in assembling sequences (see Table 1 ). Experimental validation on 454 pyrosequencing reads from HCV samples We first discuss the choice of parameters of the read graph and candidate sequence assembly from stpaths. Then we give statistical validation for obtained 10 most frequent quasispecies sequences. We infer quasispecies spectrum based on the read graphs constructed with various numbers n and m (numbers of mismatches allowed for superreads and overlaps corresponding to edges). We sort the estimated frequencies in descending order and count the number of sequences which cumulative frequency is 85%, 90%, and 95%. Fig. 3 reports these numbers as a percent of the total number of candidate sequences. There is an obvious drop in percentage for all three categories if we allow up to n = 6 mismatches to cluster reads and up to m = 15 mismatches to create edges. In this case, the constructed read graph has no isolated vertices. To refine assembled candidate sequences, we use all reads and parameter t varying from 80bp till 350bp, or, in the other words, mutation rate varying from 1.75% up to 8% per sequence (which is in the range observed in [34] ). Out of 3207 max-bandwidth paths, we obtain as much as 938 distinct sequences (t = 80) and as low as 755 sequences (t = 350) for different values of t [80; 350]. The neighbor-joining tree for the most frequent 10 candidate sequences obtained by ViSpA and ShoRAH (see Fig. 4 ) reminds a neighbor-joining tree for HCV quasispecies evolution. Additionally, the most frequent candidate sequence found by ViSpA is 99% identical to one of the actual ORFs obtained by cloning the quasispecies. The quasispecies sequence is considered found if one of candidate sequences matches it exactly (k = 0) or with at most k (1 or 9) mismatches. All methods are run 100 times on 10% -reduced data. For the i-th (i = 1, .., 10) most frequent sequence assembled on the whole data, we record its reproducibility, i.e., percentage of runs when there is a match (exact or with at most k mismatches) among 10 most frequent sequences found on reduced data. "Reproducibility: Max" and "Reproducibility: Average" report respectively maximum and average of those percentages." Figure 3 Percentage of candidate sequences which cumulative frequency is 85%, 90%, and 95%. The values on x-axis corresponds to the number of allowed mismatches during read graph construction. n_m means that up to n mismatches are allowed in superreads and up to m mismatches are allowed in edges. Viral sequences containing internal stop codons are not viable since the entire HCV genome consists of a single coding region for a large polyprotein. So the number of reconstructed viable sequences can serve as an accuracy measure for quasispecies assembly. Out of 10 most frequent sequences reconstructed by ViSpA, only 3 are not viable while ShoRAH is able to reconstruct only one viable sequence. This sequence has 99.94% similarity with the ViSpA's fourth most frequent assemblies. Both methods returned similar frequency estimations for this sequence: 0.017% (ShoRAH) and 0.019% (ViSpA). Both ShoRAH and ViSpA (n = 6, m = 15) are run on eight 2.66GHz-CPUs with 8M cache. They take around 40 minutes to assemble sequences and estimate their frequencies. Smaller value of n increases ViSpA's runtime since its bottleneck (candidate sequences assembling) is proportional to the number of reads times number of paths. Indeed, smaller value of n results in larger number of superreads in built read graph, thus, in larger set of candidate paths. For example, ViSpA runs 90 minutes for n = 2, m = 2. The plot on Fig. 5 shows validation results for 10 most frequent quasispecies sequences with respect to EM estimations assembled on Data1 by ShoRAH and ViSpA (n = 6, m = 15, and t = 120). Repeatedly, 100 times we have deleted randomly chosen 10% of reads and run both methods on each reduced read instance to reconstruct quasispecies spectrum. The plot reports the percentage of runs when each of 10 most frequent sequences assembled on Data1 are reproduced among the 10 most frequent quasispecies Figure 4 The neighbor-joining phylogenetic tree for 10 most frequent HCV quasispecies variants on a 5,205bp-long fragment obtained by ViSpA and ShoRAH. Sequences are labeled with software name and its rank among 10 most frequent assembled sequences. Percentage of runs when the i-th most frequent sequence is reproduced among 10 most frequent quasispecies assembled on the 10%-reduced set of reads. The i-th point at x-axis corresponds to the i-th most frequent sequence assembled on the 100% of reads. No data are shown for the sequences that are reproduced less than 5% of runs. inferred on the reduced instances with no mismatches (k = 0), or with k = 1, 2, 5 mismatches. For example, for k = 0 ShoRAH repeatedly (35% of times) reconstructs only the third most frequent sequence while ViSpA reconstructs 7 sequences in at least 15% times, and the most frequent sequence is reconstructed 40% times. This plot shows that the found sequences are pretty much reproducible for ViSpa. In order to compare ViSpA and ShoRAH, we run both of the methods on HIV dataset, used in the first experiment in [27] . As said above, we do not preprocess reads with respect to its 454 quality score, and it can explain poorer performance of ShoRAH. Indeed, ShoRAH correctly infers only 2 quasispecies sequences with at most 4 mismatches: one assembly has 3 mismatches with real quasispecies sequence, and the other has 4 mismatches. ViSpA correctly reconstructs 5 quasispecies with at most 2 mismatches (3 of them among 10 most frequent assemblies): two sequences are inferred without any mismatches (one is among 10 most frequent assemblies), one assembly has 1 mismatch with real quasispecies sequence (and it is among 10 most frequent assemblies), and the rest sequences have 2 mismatches (one is among 10 most frequent assemblies). The assemblies correspond to a viable protein sequences. If ViSpA is applied to ShoRAH-corrected reads, it can successfully infer three real quasispecies without any mismatches. In this paper, we have proposed and implemented ViSpA, a novel software tool for quasispecies spectrum reconstruction from high-throughput sequencing reads. The ViSpA assembler takes into account sequencing errors at multiple steps, including mapping-based read preprocessing, path selection based on maximum bandwidth, and candidate sequence assembly using probability-weighted consensus techniques. Sequencing errors are also taken into account in ViSpA's EM-based estimation of quasispecies sequence frequencies. We have validated our method on simulated error-free reads, FlowSim-simulated reads with sequencing errors, and real 454 pyrosequencing reads from HCV and HIV samples. We are currently exploring extensions of ViSpA to paired-end reads; the main difficulty is selection of pair-aware candidate paths. We also foresee application of ViSpA's techniques to the analysis of high-throughput sequencing data from microbial communities [23] and ecological samples of eukaryote populations [35] . The ViSpA source code is available at http://alla.cs.gsu. edu/~software/VISPA/vispa.html. Additional file 1: Supplementary Materials. The file contains derivation of edge cost formula (2) and EM algorithm, example of read graph construction and analysis of 454 pyrosequencing data.

Ebola Virion Attachment and Entry into Human Macrophages Profoundly Effects Early Cellular Gene Expression

Zaire ebolavirus (ZEBOV) infections are associated with high lethality in primates. ZEBOV primarily targets mononuclear phagocytes, which are activated upon infection and secrete mediators believed to trigger initial stages of pathogenesis. The characterization of the responses of target cells to ZEBOV infection may therefore not only further understanding of pathogenesis but also suggest possible points of therapeutic intervention. Gene expression profiles of primary human macrophages exposed to ZEBOV were determined using DNA microarrays and quantitative PCR to gain insight into the cellular response immediately after cell entry. Significant changes in mRNA concentrations encoding for 88 cellular proteins were observed. Most of these proteins have not yet been implicated in ZEBOV infection. Some, however, are inflammatory mediators known to be elevated during the acute phase of disease in the blood of ZEBOV-infected humans. Interestingly, the cellular response occurred within the first hour of Ebola virion exposure, i.e. prior to virus gene expression. This observation supports the hypothesis that virion binding or entry mediated by the spike glycoprotein (GP(1,2)) is the primary stimulus for an initial response. Indeed, ZEBOV virions, LPS, and virus-like particles consisting of only the ZEBOV matrix protein VP40 and GP(1,2) (VLP(VP40-GP)) triggered comparable responses in macrophages, including pro-inflammatory and pro-apoptotic signals. In contrast, VLP(VP40) (particles lacking GP(1,2)) caused an aberrant response. This suggests that GP(1,2) binding to macrophages plays an important role in the immediate cellular response.

Zaire ebolavirus (ZEBOV) is a member of the family Filoviridae within the order Mononegavirales [1] . It was discovered in 1976 in what is now the Democratic Republic of the Congo [2] as the etiological agent of a severe human viral hemorrhagic fever known as Ebola Hemorrhagic Fever (EHF). Infection with ZEBOV typically results in a rapidly fatal illness associated with high-level viremia, lack of an effective immune response, drastic lymphopenia, a severe coagulation disorder including disseminated intravascular coagulation and limited hemorrhages, widespread focal tissue necroses, systemic shock and multiorgan failure (reviewed in detail in [3] ). While the pathogenesis of ZEBOV infection has been relatively well described in experimental animals [4, 5] , only a few studies were reported that shed light on the molecular events following infection in humans. Unfortunately, these studies are partially contradictory. For instance, higher serum cytokine concentrations (IFN-a, IFN-c, IL-2, IL-6, and TNF-a) were measured in seven fatally infected patients compared to two survivors in one study, suggesting that a hyperactive immune responses may contribute to fatal outcome [6] . Other studies, describing the responses of eight fatally infected patients and four survivors, did not reveal significant concentration differences of IFN-a, IL-2, and TNF-a. On the other hand, that study suggested that fatal infections are due to generalized immunosuppression, including decreased IFN-c, IL-2, and IL-4 concentrations, lymphocyte apoptosis, and diminished IgG synthesis [7, 8, 9, 10] . The largest study to date included 42 fatally infected patients and 14 survivors. Hypersecretion of proinflammatory cytokines, chemokines and growth factors (IL-1b, IL-1RA, IL-6, IL-8, IL-15, IL-16, CXCL1 (GROa), CXCL10 (IP-10), eotaxin, M-CSF, MIP-1a, MIP-1b, MCP-1, MIF) and decreased concentrations of T lymphocyte-derived cytokines (IL-2, IL-3, IL-4, IL-5, IL-9, IL-13) concomitant to apoptotic loss of CD4 and CD8 T lymphocytes were typical for fatal cases [11] . Unfortunately, all these data reveal only the extent of homeostatic disarray in ZEBOV-infected individuals, but not its origin or genesis. It is therefore important to measure the responses of individual human cell types to infection, ideally in chronological order of their infection in vivo. Mononuclear phagocytes are very early, if not initial, targets of ZEBOV in humans and experimentally infected animals [12, 13, 14, 15] . In vitro, human and nonhuman primate macrophages are highly susceptible to ZEBOV infection with subsequent robust virus production [16, 17] , suggesting they may be the major source of the high viremia observed during the critical stages of infection. Studies performed in vitro are also strongly indicative that macrophages play a major role in inducing cytokine/chemokine dysregulation. For instance, human monocytes and macrophages infected with ZEBOV react with increased expression of MCP-1, CXCL1, IL-1b, IL-6, IL-8, MIP-1a, RANTES and TNF-a [16, 18] . Previous studies revealed that similar increased levels of expression of some of these cytokines were triggered by incubation of human macrophages with Ebola virus-like particles (VLPs) consisting of the ZEBOV matrix protein VP40 and the spike glycoprotein (GP 1,2 ) [19] or with UV-inactivated ZEBOV [16] . These findings indicated that virus replication might not be required for the activation of macrophages. The aim of the current study was to determine the gene expression profiles of human macrophages exposed to infectious Ebola virions using DNA microarray technology (reviewed in [20, 21, 22, 23, 24] ) and therefore elucidate virus-host interactions. Here, we determine the initial responses of human macrophages to Ebola virion exposure. In a first set of experiments, DNA microarray analysis was performed to determine gene expression profiles of human macrophages 1 h and 6 h after in vitro exposure to Ebola virions in comparison to mock-exposed cells. In parallel, macrophages were treated with LPS to assess the responsiveness of the cells and to compare the response to different stimuli as genes responding to both virions and LPS might highlight possible response pathways. In a second set of experiments, we distinguished between responses induced by virion binding/entry and responses that require virus gene expression, cellular signals occurring after virion entry by exposing macrophages to Ebola VLPs. We found that Ebola virions exposure, as well as exposure to VLPs can trigger most of the detected changes after 1 h of exposure, and thus independent of virus replication. Primary human macrophages were obtained from two sources. For the first set of experiments, fresh elutriated primary human monocytes from three donors (D1, D2, D3) were purchased from Advanced Biotechnologies, Columbia, MD. For the second set of experiments, primary human monocytes from three donors (Poietics CD14 + , untreated 2W-400 series) (D4, D5, D6) were purchased from Cambrex Bio Science Walkersville (Walkersville, MD). For differentiation of monocytes into macrophages, cells were cultivated in RPMI 1640 (Invitrogen, Carlsbad, CA) containing 20% heat-inactivated human AB serum (Sigma-Aldrich, St. Louis, MO), penicillin (100 U/ml), streptomycin (100 mg/ml), and L-glutamine (2 mM). Human embryonic kidney (HEK) 293T epithelial cells and grivet (Chlorocebus aethiops) kidney epithelial Vero E6 cells (ATCC, Rockville, MD) were maintained in DMEM (Invitrogen, Carlsbad, CA) containing 10% heatinactivated fetal bovine serum (Invitrogen, Carlsbad, CA), penicillin (100 U/ml), streptomycin (100 mg/ml), and L-glutamine (2 mM). All cells were incubated at 37uC in a humidified 5% CO 2 environment. The Mayinga strain of Zaire ebolavirus (ZEBOV), isolated in 1976 [2] , was used for all infections, which were performed under biosafety level 4 conditions at the National Microbiology Laboratory of the Public Health Agency of Canada in Winnipeg, Manitoba. Prior to use, virus stocks were propagated in Vero E6 cells and clarified by centrifugation at 3,000 g for 10 min at 4uC. Supernatants were then layered on TNE buffer (20 mM Tris [pH 7.5], 0.1 M NaCl, 0.1 mM EDTA) containing 20% sucrose and spun at 28,000 rpm at 4uC for 2 h by using an SW28 rotor with a Beckman Optima L-70K ultracentrifuge. The virion pellet was resuspended in RPMI 1640 and titers were determined by plaque assay as previously described [25] . As a control, supernatant from mock-infected Vero E6 cells was purified and quantified by the same procedures. Generation and purification of virus-like particles (VLPs) were performed as described elsewhere [19] . In brief, ZEBOV VLP VP40-GP and VLP VP40 were generated by transient transfection of HEK 293T cells with plasmids encoding ZEBOV VP40 and/or ZEBOV GP 1,2 [19] and quantitated by electronmicroscopic particle counting [26] and a DC protein assay (BioRad, Mississauga, Ontario). Electron-microscopic evaluation of VLPs was performed on a Phillips CM100 microscope with low dose software and Compustage attachments. Negative staining was performed on formvar carbon-coated copper grids (Electron Microscopy Sciences, Hatfield, PA). Purified VLP solution (13 ml) was exposed to a freshly glow-discharged grid for 2 min, and the grid then transferred to a drop of 1% sodium silicotungstate (pH 7.5) for Ebola virus causes a severe hemorrhagic fever syndrome in man with high case-fatality rates. Following infection, monocytes and macrophages are among the first cells targeted by the virus. These cells respond by increasing expression of inflammatory cytokines and chemokines that contribute towards pathogenesis. In order to more thoroughly characterize the host response to Ebola infection, primary human macrophages were infected with Zaire ebolavirus and samples harvested for transcriptional changes after 1 or 6 hours post infection. Whereas previous studies have analyzed a relatively small subset of host genes, this study examined the transcriptional profile of over 10,000 genes and employed rigorous pathway analyses to the datasets. Ebola virus was found to significantly regulate the expression of over 88 host genes. These changes occurred within the first hours of infection. Subsequent experiments demonstrated that virus replication was not necessary for activation. Indeed, noninfectious virus-like particles expressing the ebolavirus glycoprotein and matrix proteins were sufficient stimuli to induce activation. Prior to use, virion stocks, mock stocks, VLPs, and media were analyzed for endotoxin presence using the Limulus amebocyte lysate test (BioWhittaker, Walkersville, MD). Six days after seeding, growth medium of human macrophages was replaced with fresh RPMI 1640 containing 2% human AB serum, penicillin (100 U/ml), streptomycin (100 mg/ml), and Lglutamine (2 mM). Cells were then incubated for one day at 37uC in a humidified, 5% CO 2 environment. Cells from donors D1, D2, and D3 were infected with either mock virus preparation or ZEBOV at a multiplicity of infection (MOI) of 10. Cells of donors D4, D5, and D6 were infected with ZEBOV at an MOI of 100, mock virus preparation, ,100 particles/cell VLP VP40-GP , ,100 particles per cell VLP VP40 , mock VLP (plasmid only), ,100 latex particles/cell, or 10 ng/ml of lipopolysaccharide (LPS). Cell supernatants were removed from the cells 1 h or 6 h post infection, and RNA from cells was purified using the RNeasy Mini Kit (QIAGEN, Valencia, CA) according to the manufacturer's instructions. The analysis of purified RNA from 10 MOI was analyzed by DNA microarrays as outlined below. The RNA obtained after infection with 100 MOI was subjected to quantitative real-time RT-PCR analysis only. Genechip technology from Affymetrix (Santa Clara, CA) was used to study the transcriptional activity of human genes using the human GeneChipH array HG-U95Av2 with a total of 12,626 probes representing approximately 10,000 full-length genes (Affymetrix Technical Note to Human Genome U133 Genechip set). Normalized signal values were generated from image raw data and used to calculate p values indicating significance levels for signal strength (absent or present call) and log 2 ratio values in comparison files (change, increase, decrease calls) using the Affymetrix Microarray Suite 5.0 (MAS 5.0). Analysis was performed on these normalized data with a reduced set of genes after removing all genes that were absent on all arrays. The remaining data included 8,861 probe sets. Additional data reduction was achieved by excluding all genes that were identified as 'no change' (NC) in all six experiments when comparing Ebola virion-exposed with mock-exposed cells, respectively. The remaining dataset included 2,025 upregulated or downregulated genes. An analysis of variance (single factor ANOVA) and determination of the correlation coefficient (R 2 ) of changes in gene expression levels between ZEBOV-treated and VLP-treated cells was performed using Microsoft Office Excel. Array data were displayed on MA plots ( Figure S1 ), i.e. on a scatter plot showing the correlation between the average log 2 intensity versus log 2 ratio for a ZEBOV-treated versus mock-treated pair of cells. Log intensity was calculated as K (log(virion)+log(mock)). Genes unchanged between test and control have a log 2 value of zero; downregulated genes have negative values; and upregulated genes have positive values. Hierarchical clustering was performed using Stanford's GeneCluster and displayed with the TreeView program [27] . Clustering selection used average linkage clustering with the correlation uncentered. InnateDb (www.innatedb.com) is a publically available resource which, based on levels of either differential gene expression, predicts biological pathways based on experiment fold change datasets [28] . Pathways are assigned a probability value (p) based on the number of genes present for a particular pathway as well as the degree to which they are differentially expressed or modified relative to a control condition. For our investigation input data was limited to the subset of 2,025 genes identified above. Additionally, functional networks were created using Ingenuity Pathway Analysis (IPA) software (Ingenuity Systems, Redwood City, CA). Those genes with known gene symbols and their corresponding expression values were uploaded and mapped to their corresponding gene objects in the IPA Knowledge Base. Networks of these genes were algorithmically generated based on their connectivity and assigned a score. Genes are represented as nodes, and the biological relationship between two nodes is represented as an edge (line). The intensity of the node color indicated the degree of up-or down-regulation. Genes in uncolored notes were not identified as differentially expressed in our experiment and were integrated into the computationally generated networks on the basis of the evidence stored in the IPA knowledge memory indicating a relevance to this network. Reverse transcription and subsequent quantitative real-time polymerase chain reaction (qPCR) were performed as described previously [19] . Strand-specific qPCR was performed using strand-specific primers for the reverse transcriptase reaction. As controls for non-specific or self-priming events, control reverse transcriptase reactions lacking primer were performed in parallel. Relative amounts of different strands were determined by normalizing against the house-keeping gene GAPDH and by subtracting the amounts of PCR product resulting from selfpriming from strand-specific products. Relative quantification was performed using the comparative CT method (Applied Biosystems User Bulletin #2, Dec. 11, 1997). DNA microarray technology was used to determine the initial response of human macrophages exposed to Ebola virions. Total RNA isolated from primary human macrophages of three donors (D1, D2, D3) at 1 h and 6 h after in vitro exposure to purified Ebola virions was compared to RNA from mock-exposed cells derived from the same donors. The time points were chosen as virus gene expression does not occur within one hour of cell-virion contact, whereas it will have commenced five hours later while ZEBOV replication is still absent (see below). A total of 12 HG-U95Av2 GeneChip microarrays were analyzed to determine differences in cellular gene expression levels between Ebola virion-exposed and mock-exposed macrophages from donors D1-D3 and also compared to the responses of LPS-treated cells. Genes detected as absent (p#0.05) on all 12 arrays were removed from a total of 12,606 probe sets prior to ratio analysis resulting in a reduced number of 8,861 probe sets. The log 2 ratio was plotted as a function of the average log 2 intensity of Ebola virion-exposed versus mock-exposed samples to test for hybridization quality and for the extent of variation among biological replicates (cells from the three donors). The MA plots for both the 1 h and 6 h time points demonstrate a low variability among donors and good overall reproducibility with the mean log 2 being zero over the EBOV-Induced Changes in Macrophage Gene Expression www.plosntds.org entire signal intensity range ( Figure S1 ). The low overall variability among donors was corroborated quantitatively by an analysis of variance of the signal strength. A single factor analysis of variance (ANOVA) indicated no significant difference among the 12 arrays (p = 0.83), confirming that the expression of the majority of cellular genes was not affected by exposure to Ebola virions. Genes responsive to Ebola virion exposure were identified by data reduction, i.e. by exclusion of all genes that were identified as 'no change' (NC, p#0.05) for all conditions tested. The resulting 2,025 probe sets (genes) were characterized by at least one significant change in one donor at either 1 h or 6 h post infection. ANOVA indicated that differences in expression patterns among these probe sets were statistically significant (p = 3610 26 ). The 2,025 genes identified were then screened for patterns of cellular expression changes that could be biologically relevant. Two types of selection criteria were designed to determine whether expression of a cellular gene was significantly affected by Ebola virion exposure, threshold-based criteria and trend-based criteria. First, thresholds for fold-change, p-values, and signal strength were established. Specifically, a minimum of a 2-fold-difference in cellular gene expression levels in infected macrophages from at least two of the three donors in at least one of the two time points and p-values,0.05 were defined as pertinent ( Figure 1 ). Second, it was acknowledged that in the case of most genes the extent of changes in gene expression required for a biological impact are unknown. For instance, some genes might respond to virus infection with only minor changes in expression levels that may still be biologically relevant. Vice versa, rather dramatic changes may prove to be biologically irrelevant. Consequently, trend selection criteria were established to screen for a common tendency or direction of changes in cellular gene expression levels. Only changes with an associated p-value of #0.01 were considered, and no unique cutoff value for fold-change was specified. Accordingly, changes of less than 2-fold were accepted if the direction of change in macrophages from all donors was the same (all increased or all decreased). The two resulting gene sets were analyzed by hierarchical clustering (Stanford Cluster software) and the resulting clusters were visualized using TreeView (Figures 1 & 2) . Application of the threshold-based selection criteria identified 205 cellular genes whose expression changed upon Ebola virion exposure ( Figure 1 ). Clustering of these 205 genes revealed two subclusters. Gene subcluster 1 was characterized by extensive expression variability across donors and time points. Inspection of transcript levels (signal strength) revealed very low signals, which may be a reason of the apparent variability among biological replicates. In contrast, many genes in subcluster 2 show consistent responses among all three donors in at least one time point. However, there are some genes that show similar responses in cluster 1. Since the overall variability in subcluster 1 could be due to low signal strength the dataset of 205 genes was subjected to the additional requirement that of the two compared signal levels the higher one (i.e. mock-signal for upregulated genes, Ebola-signal for downregulated genes) has a minimum signal value of 100 (signal threshold). Fifty-one genes remained when this added threshold criterion was applied. Most of them were located in subcluster 2 and are listed in Table S1 (threshold selection). The majority of the 51 genes were upregulated at 1 h post infection, at 6 h, or at both time points. The second set of criteria, designed for selecting a trend of responses rather than assuming a threshold for fold-change, yielded a total of 66 probe sets (63 genes) (see Figure 2 and Table S1 , trend selection). Approximately half of them were characterized by a fold-change of $2 and were also identified after the first set of selection criteria was applied. A total of 88 genes were identified after either criterion was applied. Among those 88 genes from threshold and trend analysis (Table S1 ), 26 fulfilled both selection criteria. Twenty-one of these genes were characterized by altered expression levels following exposure to Ebola virions in the same direction as after treatment with LPS. Interestingly, the expression levels of cellular genes identified by applying the trend analysis (Table S1 & Figure 2 ), all changed at the 1 h time point. Importantly, this set includes genes whose elevated expression was previously associated with human Ebola virus disease, such as the genes encoding CXCL1, IL-1b, IL-6, IL-8, and TNF-a [6, 11] , as well as genes previously unknown to play a role in Ebola virus disease, such as those encoding CCL-20, COX-2, IL-15 receptor a, phosphodiesterase 4B, and t-Pa. Expression levels of almost all identified genes were upregulated, with the exception of genes encoding GNA13, CREM, and LILRB2 at the 1 h time point and MRPS6, GADD45, and DUSP2 at the 6 h time point. Pathway over-representation analysis of gene expression data Recently, the integration of bioinformatics to complex biological data sets has provided a network-based approach for delineating the host response. As cellular responses are mediated through the selective activation or repression of signaling pathways we sought to integrate our gene expression data into functional signaling networks. Functional networks were created from our biological data sets using Ingenuity Pathway Analysis (IPA) (Figure 3 ). Genes belonging to these functional networks were related to cell-to-cell signaling and interactions, hematological system development and function, immune cell trafficking, inflammatory response and cell movement. To expand on the biological significance of this analysis, and identify individual signaling pathways modulated following infection, we performed pathway over-representation analysis (ORA) with the online software tool InnateDB (www. innatedb.com) [28] and analyzed the 1 h and 6 h Ebola-infected vs. mock-infected comparative data sets. Integrated data was limited to the 2,025 genes identified above and fold-changes .1.5 and associated p-values.0.01 were chosen as parameters for pathway ORA. The resultant differentially regulated pathways with pvalues,0.1 are presented in Tables 1 and 2 . Pathway ORA of the data set for the 1 h Ebola-infected vs. mockinfected treatments identified a number of pathways directly related to activation of the G-protein coupled receptor pathway ( Table 1) . As chemokines and inflammatory mediators activate GPCR signaling this correlates with previous investigations demonstrating increased secretion of cytokines and chemokines following Ebola virus insult [16, 18] . It was also demonstrated that many of the genes identified as central nodes in the IPA functional networks (IL-6, IL-10, IRF-7, etc.) also occupied central positions in the signaling pathways identified with InnateDB (heterotrimeric GPCR signaling pathways). Indeed, the upregulation of pathways related to interleukin (IL)-2, IL-12, IL-23 and IL-27 signaling pathways were also identified as being differentially upregulated in Ebola-infected cells as compared to the mock-infected treatment (Table S1 ). Further, GPCR-related signaling pathways (cytokine-cytokine receptor interaction pathway) and the Jak-Stat signaling pathway were also upregulated during the immediate response to infection. Interestingly, pathways related to fibrinolysis (dissolution of fibrin clot pathway; fibrinolysis pathway) were also upregulated by Ebola infection as compared to the mockinfected control (Table 1) . Previously, dysregulation of the fibrinolytic system was recognized in Ebola-infected macaques [13] . A limited number of downregulated pathways were identified in our analysis as being significantly downregulated at 1 h post-infection and were largely related to metabolic processes as well as inhibition of the negative regulation of GPCR signaling (Table 1) . In contrast, differentially upregulated signaling responses at the 6 h time point were largely related to cell adhesion (syndecan-1-mediated signaling events), metabolism (prostaglandin and leukotriene metabolism; arachidonic acid metabolism; steroid hormone metabolism) and cytokine/chemokine signaling (cytokine-cytokine receptor interaction; chemokine signaling pathway; chemokine receptors bind chemokines) ( Table 2) . Interestingly, there was a large increase in the number of downregulated pathways at the 6 h time point as compared to the 1 h time point. Multiple pathways belonging to transcription and nucleotide/nucleoside metabolism were identified as being significantly downregulated in Ebola-infected cells as compared to the mock-infected controls 6 h post-infection ( Table 2) . Comparison of the differentially regulated pathway ORA data sets between the 6 h Ebola-infected vs. mock-infected and LPSstimulated vs. mock-stimulated treatments demonstrated minimal overlap of upregulated or downregulated pathways (data not shown). Whereas LPS stimulation resulted in the upregulation of pathways largely related to TNF-a, Toll-like receptor (TLR) or apoptosis signaling pathways these were not identified in the Ebola-infected samples. Thus, this likely indicates that the pro-inflammatory response to Ebola may be largely repressed by a yet unidentified mechanism at early time points following infection. Real-time RT-PCR quantification of changes in cellular gene expression levels DNA microarrays are sensitive and specific in identifying regulated transcripts, but the technique commonly leads to underestimation of the degree of fold-change in gene expression. Comparison of the fold-changes in gene expression determined by DNA microarrays to a standardized quantitative real-time RT-PCR demonstrated that the underestimation error was larger when the fold-change differences were higher, corresponding to the finding that the quantitative range of real-time RT-PCR is larger than of DNA microarrays [29] . Consequently, real-time RT-PCR was performed for a subset of 16 selected genes that fulfilled at least one of the selection criteria to verify the results obtained by DNA microarray analysis. Some of the selected genes served as controls and are known to be influenced by ZEBOV infection, however, not necessarily at these early time points (TNF-a, IL-1b, IL-6, IL-8, CXCL1, RANTES, IL-10). Some genes were selected according to their potential relevance in infection due to their known functions and roles in modulation of immune response, viral infections or signaling pathways (hCOX-2, CCL20, 4-1BB, CSF-1, G-CSF, INDO, RSAD2, t-PA, DDX5).Verification was achieved when the direction of gene expression levels was identical and when the expression level changes determined by real-time PCR were equal or stronger than those determined by DNA microarray analysis. Figure 4 illustrates the comparison of the results of microarray analysis and real-time PCR quantifications over both time points. Real-time PCR confirmed the direction of changes in gene expression levels determined by microarray analysis and verified that most changes in cellular gene expression already occurred at the 1 h time point. In particular, upregulation of genes known to play an important role in Ebola hemorrhagic fever (CXCL-1, IL-1b, IL-6, IL-8, and TNF-a [6, 11] ) was verified, although some variation was seen among donors. The observation that most cellular responses occurred within 1 h after exposure to Ebola virions suggested that virion binding/ Figure 5 ) confirmed that at 1 h cRNA amounts were below that of genomic RNA, whereas after 6 h cRNA was greatly increased. LPS served as a positive control for the responsiveness of the macrophages used in the virion exposure studies. A comparison of the effects of Ebola virion exposure versus LPS treatment revealed that the majority of genes responded similarly (Supplemental Table 1 ), further substantiating the hypothesis that the observed responses with Ebola virions were triggered by binding and/or entry. To further investigate which changes in expression levels in macrophages occur due to virion binding/entry, macrophage responses following exposure with infectious virions were directly compared to exposure to Ebola virus-like particles (VLPs) lacking viral genetic material and containing only the ZEBOV matrix protein VP40 and the ZEBOV spike glycoprotein GP 1,2 (VLP VP40-GP ) (see also [19] ). GP 1,2 is the only known surface-exposed structural component of Ebola virions and contains the cell-surface receptor-binding domain [30, 31] . We therefore hypothesized that if virion binding is responsible for the observed macrophage responses, then these effects would be mediated by GP 1,2 . To evaluate the role of GP 1,2 , VLPs consisting of VP40, but lacking GP 1,2 (VLP VP40 ), were tested in parallel with VLP VP40-GP . VLP VP40 particles were shown previously to be morphologically similar to the GP 1,2 -containing VLPs [32, 33] . LPS was again used as a control for macrophage responsiveness and also to directly compare the LPS-induced responses to those of Ebola virion and VLP exposures. Additionally, cells were exposed to latex particles, which were used in approximate equal quantity to VLPs. Ebola VLP preparations were tested by negative-stain electron microscopy for authentic appearance and presence (VLP VP40-GP ) or absence (VLP VP40 ) of GP 1,2 (data not shown). Particles resembled those previously described [32, 33] . In addition, Ebola virion, VLP, and latex preparations were tested for endotoxin-contamination before incubation with macrophages from donors D4, D5, and D6. The endotoxin concentrations of all samples used in this study were not above those of the tissue culture media (,0.5 U/ml). To determine whether the quantitative differences in gene expression levels between cells of donors D1-D3 observed in the first experiment were due to the MOI used, a higher MOI of Ebola virions (<100) was used for this experiment. Cells were exposed to VLPs and latex particles at a concentration of 100 particles per cell. LPS was used at the same concentration as in the first experiment (10 ng/ml), which allowed for a direct comparison of the responsiveness of macrophages used in both experiments. The results revealed that while a higher MOI did increase the overall intensity of responses compared to the first experiments (compare to Figure 4) , it did not eliminate or noticeably reduce the degree of variation of gene expression levels between donors ( Figure 6C & 6D ). This suggests that genetic variation among identical cell types of different donors may have an influence on the effect of ZEBOV infection and therefore may influence chances of survival. Since all stimuli were given in parallel to macrophages from each donor, this experiment allowed for a direct qualitative comparison of the cellular responses to the different stimuli. Cellular gene expression levels were determined by real-time RT-PCR specific for 21 genes that were selected from the list of genes established using DNA microarray analysis in the first experiment. These 21 genes included the majority of genes previously analyzed by real time PCR (Figure 4 ), as well as additional genes that were suspected or known to play a role in ZEBOV infection in vivo. The results of the real-time PCR quantification were analyzed by two-dimensional cluster analysis. The first clustering was performed to group genes according to similar behavior in expression changes (upregulation or downregulation) following a stimulus ( Figure 6A ). This first cluster was then used for clustering in a second dimension to group experimental stimuli according to similar effects they had on changes in gene expression ( Figure 6A , right cluster). Analysis revealed genes that were upregulated at the 1 h and then downregulated at the 6 h time point (GADD45, DUSP2, IL-10), genes that were first downregulated and then upregulated (IDO, ISG 45K), a large group of genes induced at both time points, and finally genes that were downregulated at both time points (TLR3, GNA13). The second clustering ordered the experimental groups according to similar responses in gene expression ( Figure 6B ) and revealed the following three major clusters: 1) 1 h exposure to Ebola virions, LPS, and VLP VP40-GP , 2) 6 h exposure to Ebola virions, LPS, and VLP VP40-GP , and 3) 1 h and 6 h exposure to VLP VP40 and latex particles. This result suggests that Ebola virions, LPS, and VLP VP40-GP caused strikingly similar responses in comparison to the responses to VLP VP40 and latex particles. This further supports the notion that binding/entry of virions mediated by GP 1,2 plays a significant role in the host response to infection. Thirteen of the 21 genes tested responded to Ebola virions, VLP VP40-GP , and LPS at both time points. TNF-a, IL-1b, IL-6 and IL-8 were also found in a separate study on the effects of VLPs on macrophages [19] and CXCL1 represents a gene known to be affected by ZEBOV infection. However, COX-2, GCH1, GM-CSF, MIP-3a, t-Pa, GADD45A, and IDO (6 h) are genes identified to play a role in Ebola hemorrhagic fever for the first time. Within the first two clusters, various subclusters were identified. One revealed that out of the 13 genes that were upregulated at both time points, 9 responded, albeit weakly, to VLP VP40 at the 1 h time point. This supports the notion that the matrix protein VP40 might also be involved in cellular interactions upon binding/entry, which may participate in triggering part of the detected responses in macrophages, mainly involved in proinflammatory signaling. Exposure to latex particles did not induce a response, indicating that the signals mentioned above were not due to non-specific cellular uptake of particles. Another subcluster differentiated macrophages treated with VLP VP40-GP versus Ebola virions or LPS and included IFN-inducible genes such as RSAD2 and IFIT2, as well as G-CSF, TF, GADD45, DUSP2, IL-10, IDO, and CD 137lig. Another sub-cluster revealed 3 genes, DUDP2, IL-10, and TLR3, which were downregulated at the 6 h time point, and one gene, GNA13, which was downregulated at both time points by Ebola virions or VLP VP40-GP but not by LPS. The current study represents the first broad analysis of initial transcriptional responses to Ebola virions of human macrophages, the primary target cells of ZEBOV [12, 13, 14, 15] . To assure accurate identification of cellular genes that responded significantly to virion exposure, DNA microarray data were scrutinized and stringently filtered by two different sets of selection criteria and statistical evaluations. In addition, the expression levels of selected genes were experimentally verified by real-time PCR analysis. Using these methods for statistical verification and assay evaluation, our study permitted the identification of a large number of genes not previously implicated in early cellular responses to ZEBOV infection. The detected changes of gene expression levels occurred within the first hours after primary human macrophages were exposed to Ebola virions in vitro. In addition, we identified genes whose expression is known to be upregulated during acute Ebola hemorrhagic fever. This not only validated the experimental approach and data screening criteria, it also demonstrated that elevated levels of gene products, such as cytokines, are already induced in primary target cells within the first hours of virion binding and entry. Lending further credence to these assertions, pathway ORA of the significantly differentially regulated genes identified from our gene expression studies demonstrated a strong upregulation of specific host signaling pathways related to cytokine/chemokine signaling during the immediate response to Ebola infection. Indeed, the upregulation of signaling pathways related to cytokine/chemokine signaling events suggests that specific innate immune responses are mounted during the acute phase of Ebola viral insult. Through multiple pathway ORA analyses we identified broad cellular functional networks that are modulated during the early course of Ebola infection and, importantly, have correlated this with specific cell signaling pathways. The identities of the individual signaling pathways modulated by Ebola infection will provide critical information regarding disease pathogenesis and as well information for the development of novel antiviral therapeutics. Some of the genes that had already been implicated in the pathogenesis of Ebola hemorrhagic fever such as IL-6 and TNF-a [6, 11] were induced after 1 h, but returned towards basal levels of expression after 6 h. Indeed, whereas LPS stimulation resulted in the activation of a large subset of TNF-a related signaling pathways at the 6 h time point there were no common pathways identified in the Ebola-infected pathway ORA. Similar results were obtained in another study, during which primary human macrophages, exposed to Ebola VLP VP40-GP for 24 h, were characterized by IL-1b, IL-6, IL-8, RANTES, and TNF-a expression that also peaked at 1 h or 6 h before receding to base levels [19] . These results contradict observations in vivo, which demonstrated upregulation of cytokines for extended periods of time [6, 7, 8, 9, 10, 11] . It is plausible that the early cytokine peak observed following high MOI infections in vitro is only achieved at a later time during in vivo infection. Another possible explanation for this difference is the duration of stimuli and thereby the duration of responses. For instance, in vivo, progeny virions are continuously produced by ZEBOV-infected cells and will therefore bind to and enter additional target cells, resulting in continuous stimuli that may maintain cytokine production and facilitate prolonged activation. Therefore, our in vitro approach most likely yielded results that are indicative of what continuously occurs in vivo. That increased cellular gene expression levels result in increased protein concentrations within 6 h was previously demonstrated in a study that used human macrophages exposed to VLP VP40-GP and ELISA measuring protein levels of IL-1b, IL-6, IL-8, TNF-a [19] . The identified cellular genes whose expression levels were altered after exposure to Ebola virions belonged to different functional categories (Table S1 ). These genes include inflammatory cytokines, molecules that regulate blood coagulation (such as t-Pa, MMP-1, and serpine 1 and 2) genes involved in stressresponse, DNA repair, cell cycle arrest, and cell adhesion. In support of these functional categories, our pathway ORA also resulted in similar functional categorization of the differentially regulated signaling pathways following Ebola virus infection. The detection of a group of genes that responded to binding/entry of Ebola virions, VLP VP40-GP , as well as to LPS raises the question whether LPS and Ebola virions share receptors on the macrophage surface. Whereas receptors for Ebola virions on target cells remain elusive, it is known that LPS induces its effects by binding to TLR4 and MD2 and CD14 co-receptors. Recent studies demonstrated that Ebola VLP VP40-GP , but not VLP VP40 , induced cytokine and SOCS1 expression in a TLR4/MD2 dependent manner both in a human monocytic cell line (THP-1 cells) and in 293T cells expressing a functional TLR4/MD2 receptor [34] . The innate immune defense is achieved by activating NF-kB and type I IFN responses. It is already known that ZEBOV suppresses the host cell antiviral response by inhibiting interferon signaling via its VP35 and VP24 proteins [35, 36, 37, 38, 39, 40, 41] , and RNA silencing via VP35 [42] . Our studies revealed only limited IFN signaling in Ebola virion-exposed macrophages compared to those exposed to LPS, indicating that IFN signaling is inhibited early upon infection. The host response to virulent pathogens is likely to fall into two categories [21] . First, there are common responses to unrelated pathogens, such as the type 1 IFN innate immune response that renders uninfected neighboring cells resistant to virus infection. Second, there are responses specific for individual pathogens. On the other hand, pathogens have evolved to counter these responses to ensure their own survival and transmission. Examples are how viruses of disparate families overcome the antiviral action of apolipoprotein B mRNA editing enzyme, catalytic polypeptidelike 3G (APOBEC3G) [43] , tripartite motif containing 5 (TRIM5a) [44] , bone marrow stromal cell antigen 2 (BST-2)/ tetherin [45, 46, 47] , or interferon induced transmembrane proteins (IFITM) [46, 48, 49] . It is important to remember that this study partially characterizes the response of humans to ZEBOV infection and that humans are not natural hosts of this virus. Therefore, host responses and virus counteractions are not in a state of equilibrium. It is therefore a fundamentally interesting question whether infected humans succumb to Ebola hemorrhagic fever because of direct effects exerted by the virus on the body or because of overbearing immune responses by the individual. Recently, various frugivorous bats have been implicated as potential filovirus reservoirs that seemingly remain unaffected by infection [50, 51] . It is therefore tempting to repeat our studies with cells from these animals to see whether their responses to Ebola virion exposure are fundamentally different. Taking together, our data indicate that the immediate responses of early cellular ZEBOV targets in the human organism derive from virion binding/entry mediated by the ZEBOV spike glycoprotein and do not require virus gene expression. The fact that many surveyed genes responded similarly to VLP VP40-GP , but not to VLP VP40 , treatment clearly supports this notion. However, the fact that some genes, such as those encoding TF or skin collagenase, were triggered only by Ebola virions but not VLP VP40-GP or LPS indicate that these genes must be influenced by factors other than virus binding or entry, such as other components packaged within Ebola virions in addition to GP 1,2 and VP40. In this regard, it is important to remember that infectious Ebola virions not only consist of seven structural proteins (NP, VP35, VP40, GP 1,2 , VP30, VP24, and L) but also contain cellular transmembrane proteins that are usurped by budding virions [52] . Figure S1 MA plot depicting the overall distribution of gene expression changes as a function of average signal intensity for human macrophages obtained from three different donors. Each data set compares gene expression levels Figure 6 . Determination of relative changes in expression levels of 21 cellular genes in primary human macrophages. Depicted are fold-changes at the 1 h and 6 h time points after exposure to Ebola virions compared to mock-exposed cells, cells exposed to purified Ebola virionlike particles (VLPs) containing VP40 and GP 1,2 (VLP VP40-GP ) or VLPs containing VP40 only (VLP VP40 ), or cells treated with mock-VLP preparation. As controls, macrophages of each donor were treated in parallel with latex beads or with 10 ng of LPS. and changes in primary macrophages exposed to Ebola virions compared to mock-exposure at 1 h (A) and 6 h (B). (TIF) Table S1 Summary of genes identified using threshold and trend analysis. 88 genes were identified from both threshold and trend analysis. Genes are grouped according to their known functions, determined by GeneOntology descriptors [53] . Many of these genes express proteins involved in apoptosis, inflammatory and acute immune responses, blood coagulation, and tissue remodeling. (DOCX)

Case-based reported mortality associated with laboratory-confirmed influenza A(H1N1) 2009 virus infection in the Netherlands: the 2009-2010 pandemic season versus the 2010-2011 influenza season

BACKGROUND: In contrast to seasonal influenza epidemics, where the majority of deaths occur amongst elderly, a considerable part of the 2009 pandemic influenza related deaths concerned relatively young people. In the Netherlands, all deaths associated with laboratory-confirmed influenza A(H1N1) 2009 virus infection had to be notified, both during the 2009-2010 pandemic season and the 2010-2011 influenza season. To assess whether and to what extent pandemic mortality patterns were reverting back to seasonal patterns, a retrospective analyses of all notified fatal cases associated with laboratory-confirmed influenza A(H1N1) 2009 virus infection was performed. METHODS: The notification database, including detailed information about the clinical characteristics of all notified deaths, was used to perform a comprehensive analysis of all deceased patients with a laboratory-confirmed influenza A(H1N1) 2009 virus infection. Characteristics of the fatalities with respect to age and underlying medical conditions were analysed, comparing the 2009-2010 pandemic and the 2010-2011 influenza season. RESULTS: A total of 65 fatalities with a laboratory-confirmed influenza A(H1N1) 2009 virus infection were notified in 2009-2010 and 38 in 2010-2011. During the pandemic season, the population mortality rates peaked in persons aged 0-15 and 55-64 years. In the 2010-2011 influenza season, peaks in mortality were seen in persons aged 0-15 and 75-84 years. During the 2010-2011 influenza season, the height of first peak was lower compared to that during the pandemic season. Underlying immunological disorders were more common in the pandemic season compared to the 2010-2011 season (p = 0.02), and cardiovascular disorders were more common in the 2010-2011 season (p = 0.005). CONCLUSIONS: The mortality pattern in the 2010-2011 influenza season still resembled the 2009-2010 pandemic season with a peak in relatively young age groups, but concurrently a clear shift toward seasonal patterns was seen, with a peak in mortality in the elderly, i.e. ≥ 75 years of age.

In 2009, the rapid spread of an emerging influenza virus, A(H1N1) of swine origin, resulted in the first pandemic of the 21 st century [1] . This pandemic influenza A (H1N1) 2009 virus has led to a limited outbreak in the Netherlands with, as in many other countries, generally mild illnesses in the majority of patients [2, 3] . The pandemic was considerably less lethal than was expected, with a low overall case fatality rate [4, 5] . Nevertheless, a considerable part of the pandemic influenza related deaths concerned relatively young persons (mainly young and middle aged adults) [3, [5] [6] [7] [8] . This is contrary to seasonal influenza epidemics, where deaths occur mainly amongst elderly aged 65 years or older [9] [10] [11] . In the Netherlands, all deaths associated with laboratory-confirmed influenza A(H1N1) 2009 virus infection had to be notified since 30 April 2009. This mandatory notification remained in place during the influenza season 2010-2011, which in Europe has been characterized predominantly by the influenza A(H1N1) 2009 virus, and to a lesser extent influenza virus type B [12, 13] . The national notification system provided detailed information of the clinical characteristics of all deaths associated with a laboratory-confirmed influenza A (H1N1) 2009 virus infection in the Netherlands. We performed a retrospective analysis of all fatalities, comparing the 2009-2010 pandemic season with the 2010-2011 influenza season, aiming to assess whether and to what extent pandemic mortality patterns concerning age distribution and underlying conditions were reverting to seasonal patterns. In the Netherlands, laboratory investigation was indicated for all hospitalised and/or deceased patients with suspected influenza A(H1N1) 2009 virus during the 2009-2010 pandemic as well as the 2010-2011 influenza season. Following laboratory confirmation of influenza A (H1N1) 2009 virus infection, name and clinical characteristics of hospitalised and deceased patients had to be reported to the municipal health service (MHS) by both the attending medical doctor and the head of the involved microbiology laboratory. The MHS entered the notifications into a national anonymous and passwordprotected web-based database, including structured questions about patient demographics and information on underlying medical conditions, treatments, clinical presentation, and admission to an intensive care unit (ICU). In the pandemic season 2009-2010, additional information on underlying conditions for deceased patients was collected by the Centre for Infectious Disease Control (CIb) of the National Institute for Public Health and the Environment (RIVM) in consultation with the MHS and subsequently added to the notification database. The notification database was used to perform a comprehensive analysis of all deceased patients with a laboratory-confirmed influenza A(H1N1) 2009 virus infection. Ethical approval was not required for this study as only anonymous data were used, and no (medical) interventions were made on human subjects. Based on available clinical data, the underlying medical conditions were classified into nine groups: no underlying disorders, respiratory disorders, immunological disorders (including haematological malignancies), neurological disorders, intellectual disability (including Down syndrome), cardiovascular disorders, kidney and/ or liver pathology, other non-specified malignancies and metabolic disorders. Distinction was made between patients with single and multiple underlying disorders. Descriptive statistics were calculated for all available clinical and epidemiological characteristics. No statistical significant differences between the two seasons were found for these variables. The mean age of the deceased patients in 2009/2010 was lower compared to that in 2010/2011, respectively 41 and 53 years (p = 0.02). Figure 1 shows the mortality rate per age group based on the total Dutch population in 2009 and 2010. During the pandemic season, the population mortality rates peaked in children aged between 0 and 15 years of age and in persons aged between 55 and 64 years. In the 2010-2011 influenza season, the first peak was considerably lower, while the second peak shifted to persons aged between 75 and 84 years. Data on clinical presentation was available for 42 of the 65 fatalities (65%) during the pandemic seasons and for 28 of the 38 (74%) during the 2010-2011 season (table 1). In both seasons, fatal cases presented mainly with respiratory symptoms (41%), including acute respiratory distress syndrome (ARDS), followed by systemic symptoms (17%). Immunological and respiratory disorders were the most commonly reported underlying medical conditions in the pandemic season, for respectively 23 (35%) and 22 (34%) of the 65 fatalities (table 2). In the 2010-2011 season, cardiovascular disorders and absence of medical underlying conditions were most common, respectively for 12 (32%) and 10 (26%) of the 38 deaths. Underlying immunological disorders were more common in the 2009-2010 compared to the 2010-2011 season (p = 0.02), while cardiovascular disorders were significantly more common in the 2010-2011 season (p = 0.005). Overall, multiple underlying conditions were reported for 27 of the 103 cases (26%). Particularly, intellectual disability (100%), cardiovascular (76%) and metabolic disorders (75%) were found in combination with other underlying conditions (data not shown). The peak in mortality rates in persons aged between 55 and 64 years observed during the 2009-2010 pandemic, shifted to older age groups in the 2010-2011 influenza season. Furthermore, the peak in mortality rates in children younger than 15 years of age decreased considerably. The decline of the peak in children might partly be explained by immunity in the youngest age groups, possibly related to high attack rates of influenza A(H1N1) 2009 virus in children during the pandemic season or to persisting vaccine-induced immunity [14] [15] [16] . Although the infection attacks rates during the pandemic season were very low in the older adults (≥ 40 years), the shift of the peak in mortality rates towards older age groups observed in our study might indicate increased circulation of the virus in the 2010-2011 influenza season in these age groups [14] . A shift of the age-specific mortality pattern similar to that observed in our study is also described for the post-pandemic seasons following the three pandemics in the 20 th century. During each of these earlier pandemics, persons younger then 65 years of age initially accounted for a high proportion of influenza-related deaths, followed by a declining proportion of deaths in the postpandemic seasons [17] . Simonsen et al. [17] hypothesised that younger persons may retain long-lasting immunity better than older persons after exposure to a new influenza virus subtype. Recent studies on risk factors for influenza A(H1N1) 2009 deaths concluded that the majority of severe pandemic cases as well as fatalities had underlying medical conditions as previously also associated with severe seasonal influenza [4, 8, [18] [19] [20] [21] [22] . Our results are in line with previous studies in which respiratory disorders and immunosuppressive conditions were frequently reported as underlying diseases [4] [5] [6] 18, 21] . Furthermore, neurological disorders have been reported to be common underlying diseases in fatal pandemic influenza cases, especially in children and young adults [5, 19] . Patients with neurological and neuromuscular disease have also been recognized as high-risk group for severe disease from seasonal influenza [23] . Our study showed a noticeable number of deceased patients (10%) with intellectual disability. Pérez-Padilla et al. [24] recently showed that Down syndrome was associated with adverse outcomes in cases of influenzalike illness (ILI) and severe acute respiratory illness (SARI) during the first months of the outbreak A (H1N1) 2009 influenza virus. All intellectual disabled patients in our study also had other chronic underlying conditions, making it impossible to assess the specific role of intellectual disability as a risk factor for fatal influenza. Although it has been reported that the A(H1N1) 2009 influenza virus caused severe illness and death in pregnant and postpartum women [25] [26] [27] , as observed for seasonal influenza, no pregnancy-related pandemic influenza deaths were notified in the Netherlands. As we noted for intellectual disorders, also for pregnancy fatalities it is important to verify whether other chronic underlying conditions are present. The relatively high number of fatalities with underlying cardiovascular disorders in the 2010-2011 influenza season might be associated with the shift of the mortality rates to elderly persons, since cardiovascular disorders are generally more common in elderly persons. Because of the relatively small numbers of fatalities, it is not possible to compare the differences in underlying conditions between the two seasons adjusted by age. There remains a possibility that fatal case ascertainment is incomplete because of underreporting and -diagnosing. Especially in patients with severe underlying diseases and elderly, the generally non-specific symptoms may not have been recognized as being caused by influenza A(H1N1) 2009 virus infection. This is reflected by the fact that pandemic influenza was reported as contributing cause of death in some patients instead of the main cause of death. Moreover, this might also partly explain the relatively high mortality in patients with underlying immunological disorders in the 2009-2010 pandemic season compared to the 2010-2011 influenza season. It is plausible that testing for influenza was more common during the pandemic season because of heightened attention, particularly in patients with severe underlying diseases like immunological disorders. To improve completeness of reporting in the hectic pandemic season, additional information on underlying conditions was actively collected where not available, which might have caused some information bias. Another limitation of this study is the lack of reliable historical records on deaths related to laboratory-confirmed influenza. Although deaths associated with laboratory-confirmed A(H1N1) 2009 virus infection were notifiable during the 2009-2010 and 2010-2011 seasons, clinical influenza diagnoses are generally not laboratory-confirmed during seasonal influenza epidemics. Nevertheless, estimates of the burden of seasonal influenza show that about 90% of influenzaassociated deaths occur in persons aged 65 years and older [9] . This is obviously different from the age specific mortality pattern seen during the 2009 and previous pandemics. The maintenance of the mandatory notification of deaths associated with laboratory-confirmed influenza A (H1N1) 2009 made it possible to compare the fatal cases during the 2009-2010 pandemic season with that during the 2010-2011 influenza season. The mortality pattern in the 2010-2011 season still resembles the pandemic season with a peak in relatively young age groups, but concurrently shows a clear shift towards the seasonal pattern, as also described for previous pandemics in the 20 th century.

A Chinese Herbal Formula to Improve General Psychological Status in Posttraumatic Stress Disorder: A Randomized Placebo-Controlled Trial on Sichuan Earthquake Survivors

Introduction. Posttraumatic stress disorder (PTSD) is accompanied by poor general psychological status (GPS). In the present study, we investigated the effects of a Chinese herbal formula on GPS in earthquake survivors with PTSD. Methods. A randomized, double-blind, placebo-controlled trial compared a Chinese herbal formula, Xiao-Tan-Jie-Yu-Fang (XTJYF), to placebo in 2008 Sichuan earthquake survivors with PTSD. Patients were randomized into XTJYF (n = 123) and placebo (n = 122) groups. Baseline-to-end-point score changes in the three global indices of the Symptom Checklist-90-Revised (SCL-90-R) and rates of response in the SCL global severity index (GSI) were the primary endpoints. A subanalysis of the nine SCL factors and the sleep quality score were secondary endpoints. Results and Discussion. Compared to placebo, the XTJYF group was significantly improved in all three SCL global indices (P = 0.001~0.028). More patients in the XTJYF group reported “much improved” than the placebo group (P = 0.001). The XTJYF group performed significantly better than control in five out of nine SCL factors (somatization, obsessive-compulsive behavior, depression, anxiety, and hostility (P = 0.001~0.036)), and in sleep quality score (P < 0.001). XTJYF produced no serious adverse events. These findings suggest that XTJYF may be an effective and safe treatment option for improving GPS in patients with PTSD.

On May 12, 2008 , an earthquake measuring 8.0 on the Richter scale hit Sichuan Province in southwestern China. According to the official data, more than 69,200 people were confirmed dead, more than 374,600 were seriously injured [1] , and at least 5 million were left homeless [2] . Recent literature shows that posttraumatic stress disorder (PTSD) and other psychological disorders such as anxiety and depression were fairly common and highly comorbid in 2008 Sichuan earthquake survivors [3] . Posttraumatic stress disorder (PTSD) is a significant public health problem [4] . About 6.8% of adults develop PTSD in their lifetimes; 3.5% have the condition in any given year [5, 6] . Approximately 10%-50% of the survivors of traumatic events such as earthquakes and tsunamis will develop chronic PTSD [7] , which often persists for years if untreated [8] [9] [10] . The disorder is characterized by flashbacks and avoidance or numbness as well as hyperarousal after experiencing, witnessing, or confronting actual or potential death, serious phy sical injury, or a threat to physical integrity [11] . In addition to these symptoms, co-morbid psychiatric disorders are extremely common. In the National Comorbidity Survey (USA), approximately 80% of individuals with PTSD also met criteria for at least one other disorder listed in the diagnostic and statistical manual of mental disorders-III (DSM-2 Evidence-Based Complementary and Alternative Medicine III) [4] . Patients with PTSD often manifest other complications such as depression, anxiety, obsessive-compulsive behavior, hostility, and paranoid ideation disorders [3, [12] [13] [14] [15] [16] . Co-morbid psychiatric disorders and related subclinical symptoms combined with core PTSD symptoms result in poor general psychological status (GPS). Selective serotonin reuptake inhibitors are the usual first level pharmacological treatment for PTSD [17] [18] [19] [20] [21] [22] . Other lines of drugs, such as benzodiazepines and monoamine oxidase inhibitors, are also commonly used [23] . However, the effects of these pharmaceuticals are not always satisfactory [23] [24] [25] , and undesirable side effects such as sleep disturbance, sexual dysfunction, and dizziness have been reported [23, [26] [27] [28] [29] . For centuries, traditional Chinese medicine (TCM) has been widely used in China and some other Asian countries for psychological disorders, and many classic herbal formulas have been used to treat such maladies [30] [31] [32] [33] [34] [35] [36] [37] [38] . Xiao-Yao-San is one of the most popular [30] [31] [32] [33] [34] [35] [36] . We developed a modified, granulated form of Xiao-Yao-San, Xiao-Tan-Jie-Yu-Fang (XTJYF), by adding additional herbs, mainly from another classic TCM formula Er-Chen-Tang for treating depression, and we studied the safety and effects of this modification in cancer patients with depression (see Table 1 ) [39] . Because we found the formula effective and observed no serious side effects, we hypothesized that XTJYF would improve GPS in PTSD patients. Patients were enrolled into this study five months after the 2008 Sichuan earthquake, between October 2008 and January 2009, through a community-based epidemiological survey of four settlements of a severely affected city, Dujiangyan. In the enrollment survey, the relationship between exposure to the earthquake and PTSD was assessed. Preliminary screening was performed in the communities by our researchers according to the DSM III for PTSD, Chinese version [40] . Eligible subjects were invited to participate in a diagnostic face-to-face or telephone interview with one of three experienced psychiatrists, each of which has at least eight years of clinical experience. Patients who met the inclusion and exclusion criteria were enrolled (see Patient Flow Chart, Figure 1 ), and our psychologists verified PTSD as the primary diagnosis of each enrollee. Inclusion criteria were age 16 or older, meeting DSM III criteria for PTSD with at least one of the nine Symptom Check-List-90-Revised (SCL-90-R) [41] subscores above the Chinese norm [42] , and being willing to be randomly assigned. Participants understood that those randomized into the placebo control group could receive XTJYF after completion of the whole trial if they wished. Exclusion criteria were past history of bipolarism, schizophrenia, or other psychotic disorders; current organic mental disorder, factitious disorder, or malingering; any past history of alcohol or substance dependence or abuse; evidence of clinically significant hepatic or renal disease or any other acute or unstable medical condition that might interfere with safe participation in the study; use of any medication with clinically significant psychotropic activity within two weeks of randomization; any cognitive-behavioral therapy during the trial; psychotherapy initiated or ending during the trial. For female patients of childbearing age, participation was contingent on a negative serum pregnancy test and a medically accepted method of contraception. Written informed consent was obtained from all patients before participation. Patients were free to withdraw from the study at any time. Clinical diagnoses, physicals, and laboratory examinations were mainly conducted in the outpatient clinic at the Air Force Sanatorium in the city of Dujiangyan by our psychologist and other investigators. The research staff collected patients' weekly feedback on their medical conditions and delivered the XTJYF or placebo through inhouse visits. The trial protocol was approved by the Ethics Committee of Shanghai Changzheng Hospital and the Air Force Sanatorium in Dujiangyan. A sociodemographic inventory and a medical history were taken, and a routine physical and laboratory examination (i.e., blood pressure, ECG, clinical chemistry and hematology tests, and urinalysis) was performed by the investigators as a baseline for future toxicology screening. Eligible patients were randomized to either XTJYF treatment or placebo control. Random numbers were generated by computer software; treatment codes were held by the chief investigator, who was isolated from patients and outcome data. The chief investigator was also responsible for distributing the XTJYF and placebo with the assistance of our research staff. Patients, research staff, and data entry clerks were blinded to treatment group assignment. Treatment compliance was assessed by package count and observation by the research staff. Treatment codes were disclosed after the entire study was completed. Interventions. All patients received 12 g packages of granulated XTJYF or placebo twice a day for eight weeks [39] and were instructed to drink the contents dissolved in warm, boiled water. Each patient completed the SCL-90-R questionnaires twice, at baseline prior to randomization and in the eighth week after the randomization, that is, at the end of this clinical trial. The SCL-90-R is a questionnaire for self-reporting psychological distress. It is widely used in patients suffering from mental diseases and for psychological evaluation of healthy individuals. The instrument is well accepted for its good internal consistency, dimensional structure, reliability, and validity [43] [44] [45] . The Chinese SCL-90, translated and validated by Wang from the English version of the SCL-90-R, was used [46, 47] . The SCL-90-R consists of 90 symptoms of distress. Patients were instructed to indicate the degree to which they had been troubled by each symptom during the preceding week by ranking the symptom from 0 to 4, with 0 being "not at all" and 4 being "extremely." The statements were classified into nine dimensions, or factors (F), that reflect various [41] . In addition, on the SCL-90-R, there are seven items not included in any of the nine factors, among which, three reflect sleep quality. Individual SCL-90-R factors have been used to evaluate the psychological condition of PTSD patients, and there is sufficient evidence to support the correlation of higher global SCL-90-R scores with the severity of a patient's core PTSD symptoms [12, [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] . During the trial, patients were closely monitored for adverse events (AEs) and worsening of symptoms. The time of onset of any observed or spontaneously reported AE, its duration and severity, any action taken, and the outcome were recorded. The original formula, Xiao-Yao-San, contains eight herbs: Gan-Cao (Radix Glycyrrhizae preparata), and Sheng-Jiang (Rhizoma Zingiberis recens). Our modification, XTJYF, contains all the herbs of the original formula, except Sheng-Jiang, plus additional seven herbs, including Fa Ban-Xia (Rhizoma Pinelliae preparatae) and Chen-Pi (Pericarpium Citri reticulatae), that are commonly used for psychological disorders (see Table 1 ). All herbal substances used in this trial are listed with the Pharmacopoeia Commission of China, 2005, and are accepted as suitable for human consumption when administered within standard dosage levels. None of these herbs is a controlled substance or an endangered species. Raw herbs were purchased from the Lei Yun Shang Pharmaceutical Company (Shanghai, China). The herbs were extracted with water, and the resulting granules were packaged by the Chinese Drug Preparation Department of Shanghai Changzheng Hospital. Levels of heavy metals and microbial and pesticide residues were carefully assessed, and all fell well within the normal range [58] . The placebo granules, purchased from Jiangsu Tianjiang Pharmaceutical Company, Ltd., were designed to resemble the XTJYF granules in taste, smell, and appearance. The placebo was composed of dextrin, sunset yellow fcf, and a sweetener; the proportion was 1200 : 1 : 7. After being tested on five independent volunteers, the placebo was deemed indistinguishable from XTJYF. XTJYF and the placebo were dispensed in identical opaque packages. 2.6. Statistical Analysis. Quantitative data was summarized using mean, standard deviation (SD), or 95% confidence interval (95% CI). Qualitative data was described using proportion, as percentages. Baseline characteristics of the two groups were compared using the two-sided chi-square test or t-test at a significance level of 0.05. Since this was a randomized, blind clinical trial, the statistical analyses for treatment effect evaluation of the primary and secondary outcomes are relatively straightforward. Baseline-to-end-point score changes in the three global SCL-90-R indices and rates of response in the GSI were computed as the primary endpoints. For defining rate of response, patients with a reduction of at least 30% from the baseline GSI score were classified as "much improved"; at least 50%, as "very much improved." Subanalyses of the baseline-toend-point score changes of the nine SCL factors and sleep quality score (the average of the scores of the three SCl-90-R items on sleep quality) were secondary endpoints. Statistical analysis on both primary and secondary outcomes was done using intention-to-treat analysis (ITT) with statistical software SPSS. Missing values in the SCL-90-R questionnaire for the patients who withdrew from the study before the eighth week were imputed using the last-observation-carried-forward method. For primary outcomes, effect sizes (for three global indices) and number needed to treat (NNT, for rate of response in the GSI), as well as the P values from two sample t-tests and chi-square tests, are reported in the treatment effect assessment. The same analytic approaches were applied to the secondary outcomes. Additionally, Fisher's exact test was used to compare the difference in dropout rate and AEs between the two treatment groups. A total of 3478 individuals were screened, of whom 820 passed the preliminary screening and 245 were finally enrolled into the study; 575 were excluded. Of these, 372 were lost to follow-up or refused enrollment; 178 did not meet the inclusion criteria; 25 met the exclusion criteria. Enrolled patients were randomly assigned to XTJYF (n = 123) or placebo (n = 122) treatment. Of these, 102 (83%) of the XTJYF group and 99 (81%) of the control group completed the whole study. Reasons for withdrawal from the study are listed separately for each treatment group in Figure 1 , and a detailed discussion on treatment tolerability is provided in Section 3.4. Table 2 shows that randomization was effective and that there were no significant differences between the two groups in baseline demographics, core clinical PTSD symptoms, or baseline SCL-90-R global indices. Even though individual SCL-90-R factor scores and sleep quality scores at baseline are not shown here, we checked all of them and founded no significant differences between the two groups. Notice that women constituted 72% of XTJYF-treated and 71% of placebo-treated patients. Ages ranged from 16 to 85; 64% were over 45. Table 3 shows the urgency of the public health needs of these earthquake-affected PTSD patients and indicates Table 4 shows that patients in the XTJYF group experienced statistically significant improvement after treatment in all three supplementary global index scores compared to the placebo group. Based on the reported effect sizes, XTJYF treatment has a moderate effect on GSI and PSDI indices and a small effect on the PST index. Our findings on the rate of response, defin-ed based on GSI score improvement, are displayed in Figure 2 ; 50% of the XTJYF patients versus 28% of those in the placebo group were "much improved," providing statistically significant evidence supporting the advantage of XTJYF over placebo at the level of 0.05 (P value = 0.001). The NNT is 4.55. Additionally, as Figure 2 shows, 20% of the XTJYF patients versus 12% of those in the placebo group were "very much improved," but this result is not statistically significant (P value = 0.12). Outcomes. The second part of Table 4 displays the treatment effects of XTJYF and placebo on the nine SCL factors and sleep quality score. The results indicate that, in comparison to placebo, the XTJYF group experienced statistically significant improvement after treatment in five of the nine SCL factors, somatization (P = 0.003), obsessive-compulsive behavior (P = 0.036), depression (P = 0.001), anxiety (P < 0.001), and hostility The original data was obtained from Jin et al. [42] . We recalculated the original data from "mean (sd)" to "mean (95% CI)" in order to make these data comparable. • The original data was obtained from Derogatis [41] . We recalculated the original data from "mean (sd)" to "mean (95% CI)" in order to make these data comparable. * Compared to the Chinese and American norms, P < 0.05. Compared to the American norms, P < 0.05. Table 4 : XTJYF treatment effect on primary and secondary outcomes. (1) Statistical analysis was done using intent-to-treat analysis (ITT) with SPSS. (2) Cohen's d effect size measure, in which an effect size of 0.2 to 0.3 is considered a "small" effect, around 0.5, a "medium" effect, and 0.8 to infinity, a "large" effect, is used here. (3) The P values come from the two sample t-tests. (P = 0.019). Based on the reported effect sizes, XTJYF treatment has a moderate effect on somatization, depression, anxiety, and hostility, as well as a small effect on obsessivecompulsive behavior, interpersonal sensitivity, and phobic anxiety. Table 4 also shows that XTJYF treatment yielded statistically significant improvement in sleep quality at the end of the study, with a P value of less than 0.001 and a moderate effect size. Overall, XTJYF was well tolerated. Compliance rate, 83% for the XTJYF and 81% for the placebo group, was reasonably high. Six in the XTJYF and five in the control group withdrew due to adverse effects, so reported AEs were similar in the two groups. The most frequently reported AEs were nausea (14.6% versus 9.0%; P = 0.24), diarrhea (10.6% versus 6.5%; P = 0.36), and malaise (10.6% versus 12.3%; P = 0.69). All AEs were minor and were determined to be unrelated to the ingestion of XTJYF. In the XTJYF group, 21 subjects dropped out (17.1%); in the placebo group, 23 did (18.9%, P = 0.74). The primary reasons cited for dropout in the XTJYF and placebo groups, respectively, were AE (4.9% versus 4.1%; P = 1); lost to Patients with a score reduction of at least 30% from the baseline SCL-90-R GSI score were classified as "much improved," and 50%, as "very much improved." * XTJYF versus placebo, P < 0.05. follow-up (1.6% versus 2.5%; P = 0.68); protocol violation (4.9% versus 3.3%; P = 0.75); lack of efficacy (3.3% versus 6.6%; P = 0.25); miscellaneous reasons, for example, disliked the taste of the herbs (2.4% versus 2.5%; P = 1). Subjects' laboratory values and vital signs were similar in the two groups. Changes in these values were minor, infrequent, and not considered clinically meaningful. In the present study, we compared our data to the Chinese norm calculated by Jin et al. [42] and to the USA norm published by Derogatis [41] . (see Table 3 ). At baseline, the nine SCL factors and three global indices were higher than the norm in these earthquake-related PTSD subjects, suggesting that earthquake-related PTSD is accompanied by poor GPS. These findings are consistent with those reported by other investigators [3, [59] [60] [61] [62] [63] . Hypothesizing that it would improve poor GPS in earthquake-related PTSD, we investigated a Chinese herbal formula, XTJYF, modified from a classic formula, Xiao-Yao-San, and found that, compared to placebo, XTJYF significantly improved all of the three global indices of SCl-90-R, and a significantly greater proportion of patients were "much improved" according to changes in GSI score. (See Table 4 and Figure 2) . A subanalysis provided a more detailed look at specific XTJYF effects on poor GPS, showing that five of the nine SCL factors and sleep quality score improved. (see Table 4 ). These findings suggest that XTJYF may globally improve GPS in earthquake-related PTSD patients, specifically in somatization, obsessive-compulsive behavior, depression, anxiety, and hostility. In addition, the formula may improve the sleep quality of the patients and appears to be safe. Although a few subjects reported gastrointestinal complaints such as nausea and diarrhea during treatment, these were probably due to the poor diet available after the earthquake; these symptoms were also frequently reported in the placebo control group. Our findings are consistent with those of our previous study on XTJYF for cancer patients with depression [39] . The results are meaningful because all five of the psychological disorders mentioned above are associated with high levels of functional and psychosocial disability in chronic PTSD patients [3, 4, 6, [12] [13] [14] [15] [16] [64] [65] [66] [67] [68] [69] , and most are reported to predict greater refractoriness to routine therapy in individuals diagnosed with PTSD [17, [70] [71] [72] [73] . For example, PTSD patients who report somatic symptoms also report higher overall PTSD symptoms [15, 64] and a higher frequency of depression [64, 74] . Patients with co-morbid PTSD and obsessive-compulsive behavior have been found to have a poorer response to cognitive behavioral therapy than those diagnosed with PTSD alone [73] . Co-morbid PTSD/depression appears to predict greater refractoriness to pharmacotherapy, greater symptom severity, lower levels of functioning and rates of recovery, and increased disability and potential of suicide [4, 6, [65] [66] [67] . Like depression, anxiety symptoms are associated with lower quality-of-life estimates and greater refractoriness to routine pharmacotherapy in PTSD patients [3, 68] . Hostility, which according to a meta-analysis of 39 studies is significantly elevated in individuals with PTSD [16] , is linked to adverse health outcomes, including cardiac death [69] . In the present study, XTJYF also appears to improve the sleep quality of these PTSD patients; sleep disturbances are among the most treatment-resistant symptoms of PTSD [75] . All of these symptoms are likely to contribute to alcohol and drug abuse [76, 77] as well as suicidal ideation [78] . The psychological mechanisms of action of Xiao-Yao-San and its modifications have been investigated. It has been reported that the formula may act on psychological symptoms by upregulating central neurotransmitters such as serotonin. Bao et al. [79] reported that Xiao-Yao-San produced antidepressant effects in a mouse model of depression by ameliorating brain cortex 5-HT and 5-HIAA content. Other mechanisms of the formula have also been reported. Yue et al. [80] reported that Xiao-Yao-San suppressed chronic stress in a rat model by up-regulating GluR2/3, the AMPA receptor subunit 2/3, which mediates the postsynaptic depolarization that initiates neuronal firing [81] , and by downregulating PICK1, a protein that interacts with C-kinase 1, which may lead to AMPA receptor anchorage [82] in hippocampal regions CA1 and CA3. Similar findings, that Xiao-Yao-San upregulates AMPA receptor subunit mRNA expression in hippocampal region CA1 and the amygdala, were reported [83] . Furthermore, Xiao-Yao-San and its modifications were reported to suppress chronic stress by maintaining the stability of hippocampal neurons [84] , inhibiting hypothalamic-pituitary-adrenocortical axis negative feedback regulation [85] , and counteracting increase of Ca 2+ concentration in hippocampal synaptosomes [86] . Based on TCM theory, seven drugs were added in our modification, mainly from another classic TCM formula, Er-Chen-Tang. According to our previous preclinical study, this modification may 8 Evidence-Based Complementary and Alternative Medicine suppress depression by up-regulating the 5-HT 1A receptor in the hippocampus in a rat model of chronic stress [87] . However, because Xiao-Yao-San and its modifications contain multiple ingredients, specific active ingredients have not been identified, and the herbal interactions within the formula have not been systematically investigated. Further investigation to elucidate the mechanisms of action of this formula is warranted. Several limitations to this study should be noted. First, our trial lacked a long follow-up assessment. This was largely due to the difficulties in following up this particular population, which consisted of earthquake survivors living in shelters with no specific address. In the patient recruitment stage, more than 45% (372 of 820) of those preliminarily screened for PTSD were lost to follow-up. Secondly, we did not include a questionnaire measuring specific PTSD core symptoms, mainly because of the low level of education in this mountain population. In our patient population, 43% had an elementary education or less and found it difficult to complete a single 90-question SCL-90-R questionnaire. However, although we did not include a specific questionnaire such as the Clinician-Administered PTSD Scale [88] or the Clinician-Rated Treatment Outcome PTSD Scale [89] to measure core PTSD symptoms, the widely used SCL-90-R captures a broader patient psychological profile than a specific PTSD questionnaire would do. Thirdly, only one dosage of XTJYF was used in this study, that used in our standard clinical practice. A higher dosage might benefit the nonresponders. Finally, more detailed information on types of trauma and the percentages of patients who suffered them should be gathered and analyzed. Despite the limitations, our findings provide preliminary support for the use of TCM in treating GPS in earthquake survivors with PTSD. TCM has been used extensively in China to treat people suffering from various diseases after disasters, for it is readily available, reasonably cheap, effective, and safe. Because of their wide usage, the production of TCM herbal products is quick and cost effective in China. Traditional Chinese herbal medicine may provide an adjuvant therapy that is safe, effective, and timely for affected populations in natural disasters such as earthquakes.

Quercetin 7-rhamnoside reduces porcine epidemic diarrhea virus replication via independent pathway of viral induced reactive oxygen species

BACKGROUND: On the base of our previous study we were observed relevant studies on the hypothesis that the antiviral activity of quercetin 7-rhamnoside (Q7R), a flavonoid, won't relate ability of its antioxidant. METHODS: We were investigated the effects of Q7R on the cytopathic effects (CPE) by CPE reduction assay. Production of DNA fragment and reactive oxygen species (ROS) induced by PEDV infection were studied using DNA fragmentation assay and flow cytometry. RESULTS: In the course of this study it was discovered that Q7R is an extremely potent compound against PEDV. The addition of Q7R to PEDV-infected Vero cells directly reduced the formation of a visible cytopathic effect (CPE). Also, Q7R did not induce DNA fragmentation. Furthermore, ROS increased the infection of PEDV, which was strongly decreased by N-acetyl-L-cysteins (NAC). However, the increased ROS was not decreased by Q7R. Antiviral activity of antioxidants such as NAC, pyrrolidine dithiocarbamate (PDTC), and the vitamin E derivative, trolox, were hardly noticed. CONCLUSIONS: We concluded that the inhibition of PEDV production by Q7R is not simply due to a general action as an antioxidants and is highly specific, as several other antioxidants (NAC, PDTC, trolox) are inactive against PEDV infection.

Many viruses are capable of inducing cell death, leading to lysis of the infected cells [1] [2] [3] [4] [5] [6] [7] . In late stages of virus infections, morphological changes, commonly known as cytopathic effect (CPE), can be microscopically observed. Virus-induced CPE is characterized by cell rounding, shinkage, deformation of nuclei and chomatin condensation. However, early death of infected cells may limit virus replication [8] . Also, apoptosis, or programmed cell death (PCD), during the late phase of viral infection has been suggested to play an important role in virus life cycle by facilitating viral progeny release and propagation [9, 10] . PCD is a process by which damaged, aged, or otherwise unwanted cells are eliminated though a series of steps that results in the destruction of their genome. The form of PCD known as apoptosis is characterized by a series of morphological changes, including nuclear condensation and fragmentation, cytoplasmic blebbing, and cell shinkage [4] . Many viruses are capable of inducing reactive oxygen species (ROS) production. Results from many studies suggest that ROS are not directly involved in the induction of apoptosis in virus-infected cells [11, 12] . On the other hand, it has been demonstrated that virus infection increases the production of superoxide anion radicals from neutrophils and macrophages infiltrated into the lung of mice [13] , while transgenic mice carrying over-expressed extracellular superoxide dismutase exhibited less severe lung injury after influenza virus infection [14] . These studies, therefore, postulated that the pathogenesis of virus infection involves not only the virus proliferation mediated apoptotic cell death in the infected cells, but also the direct ROS-induced cellular injury by neutrophils and macrophages infiltrated into the virus-infected organs. But, despite many studies, the events leading to the generation of ROS during viral infections are still unclear. In this paper, we was demonstrated the effects of quercetin 7-rhamnoside (Q7R) on production of CPE, ROS and DNA fragmentation inducted by PEDV infection and also studied the relationship of antiviral and antioxidant activity between Q7R and antioxidants. Ribavirin and sulforhodamine B (SRB) were purchased from Sigma-Aldrich (St. Louis, MO, USA). All other chemicals were a reagent grade. Q7R was isolated from aerial parts of Houttuynia cordata using a previously described method [15] . Vero (an african green monkey kidney cell line; ATCC CCR-81) was kindly provided by ATCC (American Type Culture Collection, Manassas, VA, USA). PEDV CV 777 (porcine epidemic diarrhea virus) was obtained from national veterinary research & quarantine service in Korea. Vero cells were maintained in minimal essential medium (MEM) supplemented with 10% fetal bovine serum (FBS) and 0.01% antibiotic-antimycotic. Antibiotic-antimycotic, trypsin-EDTA, FBS and MEM were supplied by Gibco BRL (Grand Island, NY). The tissue culture plates were purchased from Falcon (BD Biosciences, NJ, USAs). Virus stock was stored at -70°C until use. The antiviral activity and cytotoxicity of Q7R against viruses were determined by cytopathic effect (CPE) reduction method recently reported [15] . Also, the effect of Q7R on PEDV-induced CPE was observed by cytopathic effect (CPE) reduction method recently reported [15] . Ribavirin was used as positive, and was solublized in dimethylsulfoxide (DMSO) used as negative control. The level of intracellular ROS was measured by the alteration of fluorescence resulting from oxidation of 2', 7'-dichlorofluorescein diacetate (DCFH-DA, Molecular Probes, Eugene, OR). DCFH-DA was dissolved in DMSO to a final concentration of 20 mM before use. For the measurement of ROS, cells were treated with Q7R and other reagents for a time period indicated in the figure legends. After washing twice with cold PBS, they were incubated with 20 μM DCFH-DA at 37°C for 15 min. DCFH-DA is a stable compound that easily diffuses in to cells and is hydrolyzed by intracellular esterase to yield a reduced, non-fluorescent compound, DCFH, which is trapped within cells. The ROS produced by cells oxidized the DCFH to highly fluorescent 2', and 7'-dichlorodihydrofluorescein (DCF). The intensity of fluorescence was recorded using a flow cytometry (Becton Dickenson), with an excitation filter of 530 nm and an emission filter 575 nm. The ROS level was calculated as a ratio of: ROS = mean intensity of exposed cells: mean intensity of unexposed cells. Vero cells were seeded onto a 6-well culture plate at a concentration of 2 × 10 4 cells per well. Next day, medium was removed and the cells were washed with PBS. Then, 0.09 ml of diluted virus suspension and 0.01 ml of medium supplemented with typsin-EDTA containing an appropriate concentration of the antiviral compound were added. It was a ten-fold dilution scheme for each compound. The culture plates were incubated at 37°C in 5% CO 2 for 2 days, the cells were lysed with lysis buffer (TE buffer;10 mM Tris-HCl, pH 8.0, 100 mM NaCl, 10 mM EDTA, 0.5% SDS] and incubated for 10 min on ice, then centrifuged at 13,000 rpm for 30 min at 4°C. Cysolic DNA was extracted by phenol: chlororform (1:1) extraction of the supernatants. DNA was treated with 0.1 mg/ml Rnase A for 30 min at 37°C. The DNA was separated by agarose gel electrophoresis, and the DNA fragmentation was visualized from the digitized image of the gel as described [12] . The effect of Q7R on PEDV-induced CPE After 2 day infections of Vero cells with PEDV, Mock cells ( Figure 1A ) or cells treated with 10 μg/ml Q7R ( Figure 1C ) or ribavirin ( Figure 1B) showed typical spread-out shapes and normal morphology. At this concentration, no signs of cytotoxicity of Q7R were observed. Infection with PEDV in the absence of Q7R resulted in a severe CPE ( Figure 1D ). Addition of Q7R on infected Vero cells inhibited the formation of a visible CPE ( Figure 1F ). However, the addition of ribavirin in PEDV-infected Vero cell was impossible to prevent CPE ( Figure 1E) . Thus, the CPE of the virus infection is prevented by the presence of Q7R. To investigate the possible mechanisms of the observed antiviral effects of Q7R, we first examined whether the antioxidant property of Q7R contributed to its action. To determine the influences of PEDV replication on intracellular ROS level, Vero cells were infected with PEDV for various periods of time, and a fluorescence probe, DCFH-DA, was added to the medium prior 15 min to use a flow cytometry. As shown in Figure 2A and 2B, exposure to PEDV resulted in increased ROS production which began at 2 h post-infection and peaked at 6 h (Figure 2A and 2B) . To study the influence of antioxidant on ROS increased by PEDV infection, Q7R and NAC were used. As shown in Figure 3A and 3B, at 3 h post-infection, PEDV infection resulted in a drastic increase of intracellular ROS, which was strongly decreased by NAC but not by Q7R, according to increase of its concentration. To study the influence of antioxidants on PEDV replication, the antioxidants, NAC, trolox and PDTC were used. As shown in Figure 4A and 4B, the four compounds did not exhibit cytotoxicity at different concentrations. Antiviral activity of NAC and PDTC was hardly present. However, trolox strongly exhibited antiviral activity at 100 μg/ ml concentration, but decreased rapidly according to a dose dependent manner compared with Q7R. The effect of Q7R on the extent of DNA fragmentation resulting from PEDV infection was examined. The incubation with Q7R or ribavirin up to 100 μg/ml concentration for 48 h did not induce DNA fragmentation in mock-infected Vero cells ( Figure 5A or 5B, lanes 2-4). PEDV infection induced DNA fragmentation in Vero cells 48 h after infection in the absence of compounds ( Figure 5A or 5B, lanes 5). In the presence of Q7R, the DNA fragmentation was not induced in a dose-dependent manner ( Figure 5B, lanes 6-8) . But, DNA fragmentation was somewhat decreased when Vero cells infected with PEDV were treated with ribavirin at concentration of 1 or 10 μg/ml for 48 h. Incubation with ribavirin of 100 μg/ml did not induce DNA fragmentation. Many viruses are capable of inducing cell death, leading to lysis of infected cells [1] [2] [3] . In late stages of PEDV infections, morphological changes commonly known as CPE, microscopically observed. The morphology of Vero cells after infection with PEDV was greatly decreased from that of PEDV by addition of Q7R. However, the addition of ribavirin to PEDV-infected Vero cell proved to be impossible in preventing CPE. Viral infections such as rhinovirus, influenza virus, human immunodeficiency virus and bovine viral diarreha virus frequently result in the generation of oxidative stress in the infected cells [4] [5] [6] [7] . The events leading to the generation of ROS during viral infections are still unclear. Also, antioxidants have been shown to have Figure 1 The effects of Q7R on PEDV-induced CPE. Culture medium 6-well tissue culture plates were removed and the cells were washed with PBS. Then, 0.09 ml of diluted virus suspension and 0.01 ml of medium supplemented with typsin-EDTA containing Q7R of 10 μg/ml were added. After incubation at 37°C in 5% CO 2 for 2 days, the morphology of cells was investigated under microscope and a photograph taken. antiviral activities against a variety of unrelated viruses by alleviating the oxidative stress generated by viruses [16] [17] [18] [19] [20] . Mechanistically, it is believed that these viruses induce apoptosis by oxidative stress mediated via ROS. Interference with this pathway by antioxidants is believed to inhibit virus-induced apoptosis and thus inhibit efficient virus multiplication. In contrast, there are also reports indicating that under certain conditions compounds act as a pro-apoptotic drug [21] [22] [23] . Depending on the viral system analyzed, antioxidative compounds differ in their ability to reduce virus growth [7, 24, 25] . Flavonoids are a large class of polyphenolic compounds and Jung et al. (2003) reported that the relationship between flavonoid structure and antioxidant activity. They found that the inhibitory activities of flavonoids on total ROS are more strongly increases with the rising number of hydroxyl groups than the foavonoid glycosides on their structures [26] . Our previous study showed that quercetin 7-rhamnoside (Q7R) didn't directly interact with PEDV particles and affect the initial stage of PEDV infection by interfering with its viral mRNA production [15] . In this report, we present evidence that Q7R, but not other commonly used antioxidants, are able to protect cells from PEDV induced death. Q7R are potent agents that have been shown to be involved in a number of processes, suggesting that their antiviral effects might not be due to its antioxidant functions alone. Nevertheless, further studies are needed to verify the underlying mechanism of Q7R action in inhibiting PEDV infection. In conclusion, Q7R is an extremely potent anti-PEDV substance which reduces PEDV growth, inhibits Inhibition of PEDV replication by Q7R is independent of its antioxidant activity. Vero cells were infected with PEDV (MOI = 10) in the presence of Q7R. Relative ROS generation in Vero cells exposed to Q7R of 1, 10, 100 and 500 μg/ml for the indicated times. NAC exposed to 10 μg/ml for the indicated times. Redox-sensitive fluorescence probe DCFH-DA (20 μM) was added to the phosphate buffered-saline (pH 7.2) after harvest of Vero cells infected with PEDV at 6 h. Representative images of ROS-induced DCF fluorescence of infected cell at 6 h post-infection are shown at a flow cytometry histogram. All data represent mean values of six independent measurements ± S.D. Ctr2, control treated with DCFH-DA; NAC 10, treated with N-acetyl L-cysteine of 10 μg/ml; PEDV, infected with PEDV for 6 h; Q7R 500, treated with Q7R of 500 μg/ml. the CPE and DNA fragment of infected cells regardless of its antioxidant activity and then didn't directly interact with PEDV particles and affect the initial stage of PEDV infection by obstructing with its viral mRNA production. It will be interesting to further investigate the antiviral activity of the Q7R in preventing various PEDV-mediated injuries in in vivo pathological situations.

Use of Recombinant Adenovirus Vectored Consensus IFN-α to Avert Severe Arenavirus Infection

Several arenaviruses can cause viral hemorrhagic fever, a severe disease with case-fatality rates in hospitalized individuals ranging from 15-30%. Because of limited prophylaxis and treatment options, new medical countermeasures are needed for these viruses classified by the National Institutes of Allergy and Infectious Diseases (NIAID) as top priority biodefense Category A pathogens. Recombinant consensus interferon alpha (cIFN-α) is a licensed protein with broad clinical appeal. However, while cIFN-α has great therapeutic value, its utility for biodefense applications is hindered by its short in vivo half-life, mode and frequency of administration, and costly production. To address these limitations, we describe the use of DEF201, a replication-deficient adenovirus vector that drives the expression of cIFN-α, for pre- and post-exposure prophylaxis of acute arenaviral infection modeled in hamsters. Intranasal administration of DEF201 24 h prior to challenge with Pichindé virus (PICV) was highly effective at protecting animals from mortality and preventing viral replication and liver-associated disease. A significant protective effect was still observed with a single dosing of DEF201 given two weeks prior to PICV challenge. DEF201 was also efficacious when administered as a treatment 24 to 48 h post-virus exposure. The protective effect of DEF201 was largely attributed to the expression of cIFN-α, as dosing with a control empty vector adenovirus did not protect hamsters from lethal PICV challenge. Effective countermeasures that are highly stable, easily administered, and elicit long lasting protective immunity are much needed for arena and other viral infections. The DEF201 technology has the potential to address all of these issues and may serve as a broad-spectrum antiviral to enhance host defense against a number of viral pathogens.

The Arenaviridae family of viruses has several members that can cause viral hemorrhagic fever, an acute, often-fatal, viral syndrome characterized by intense fever, malaise, and less frequently, bleeding and neurologic manifestations. Case fatality rates of hospitalized patients suffering from arenaviral hemorrhagic fever (AHF) range from 15-30% [1, 2, 3, 4] . Arenaviruses known to cause AHF include Junín, Machupo, Guanarito, Sabiá, and Chapare in the South American continent, and Lassa and Lujo in west and southern Africa, respectively. Primary transmission of the arenaviruses from respective rodent reservoir hosts to humans occurs via exposure to contaminated excreta [5] . Person-to-person transmission can occur through contact with blood or other body fluids during the care and management of infected individuals [1, 6] . Notably, these viruses are considered a threat to national security and are classified as highest priority pathogens by the NIAID [7] . At present, the treatment of AHF is limited to ribavirin and immune plasma [8, 9] . The latter has only been proven to be effective in treating cases of Argentine hemorrhagic fever (Junín virus infection) within 8 days of disease onset. Off-label usage of ribavirin has been shown to be effective in treating Lassa fever when therapy was initiated within 6 days of the development of clinical symptoms. However, there are toxicities associated with ribavirin therapy at dosages required for efficacious use, which may contribute to the observed poor patient compliance in completing prescribed treatment regimens [10, 11] . Very limited case data using ribavirin to treat other AHFs supports the use of emergency protocols [1, 12, 13] , however the utility of ribavirin therapy remains to be seen. Interferon alpha (IFN-a) is an effective part of the host innate immune response, which can be manufactured as a recombinant human protein with broad clinical appeal [14] . Consensus (c)IFNa, also known as IFN alfacon-1 and Infergen, is a licensed, second generation IFN-a engineered to contain the most frequently occurring amino acids among the nonallelic IFN-a subtypes. Previously, we have demonstrated that cIFN-a can be used effectively alone, or in combination with ribavirin, to treat Pichindé virus (PICV) infection in hamsters [15, 16] , an experimental model of acute arenaviral disease [17] . However, while cIFN-a has clinical value, its usefulness is hindered by its short half-life and cost to manufacture. There is an initial distributive half-life of 7 minutes and a beta half-life of 2 to 5 hours [14] . The rapid systemic clearance requires frequent dosing to achieve desired therapeutic levels. Consequently, treatment can result in well-documented toxicities which include headache, depression, hair loss, fever, and malaise. In order to combat the rapid degradation, PEGylated forms of recombinant IFN-a have been introduced with half-lives that are on the order of days instead of hours, thus reducing the number of injections to once per week [18] . However, the cost to manufacture PEG-IFN-a is exceedingly high, and the PEGylation process has been shown to reduce the activity of IFN-a, thereby further increasing the production costs. To circumvent the fast decay of cIFN-a, a replicationincompetent, recombinant adenovirus type 5 (rAd5) gene delivery platform was designed to drive constitutive expression of the cIFNa gene from transduced nasal epithelial target cells. This rAd5 cIFN-a virus, called DEF201, was first developed in mice and recently shown to be active against yellow fever virus (YFV) infection in hamsters [19, 20] . The intranasal (i.n.) inoculation used in the YFV study prevents the host immune system from recognizing the Ad5 vector, thereby bypassing any possible preexisting immunity [21] . In the present study, we evaluated the use of DEF201 administered i.n. for the prevention and treatment of PICV infection in hamsters. All animal procedures complied with USDA guidelines and were conducted at the AAALAC-accredited Laboratory Animal Research Center at Utah State University under protocol 1229, approved by the Utah State University Institutional Animal Care and Use Committee. Female golden Syrian hamsters were obtained from Charles River Laboratories (Wilmington, MA) and acclimated for a minimum of 6 days prior to experimentation. They were fed standard hamster chow and tap water ad libitum. Animals were approximately 7-9 weeks old at the time of virus challenge. PICV, strain An 4763, was provided by Dr. David Gangemi (Clemson University, Clemson, South Carolina). The virus was passaged once through hamsters. Virus stocks were prepared from pooled livers harvested from infected hamsters. Virus dilutions were made in minimal essential medium (MEM), and infectious inoculum was given bilaterally in two intraperitoneal (i.p.) injections of 0.1 mL each. The recombinant adenovirus vectored cIFN-a (rAd5-huIFN-a; DEF201) and the rAd5 empty vector (rAd EV) control virus were provided by Defyrus, Inc. (Toronto, ON, Canada) at a concentration of 6610 9 and 2610 11 plaque-forming units (pfu)/ml, respectively. Both viruses were prepared in PBS for i.n. instillation in a 200 ml volume. Virus titers were assayed using an infectious cell culture assay as previously described [22] . Briefly, a specific volume of liver or spleen homogenate or serum was serially diluted and added to triplicate wells of Vero (African green monkey kidney; American Type Culture Collection, Manassas, VA) cell monolayers in 96well microplates. The viral cytopathic effect (CPE) was determined 7 to 8 days post-virus inoculation, and the 50% endpoints were calculated as described [23] . The assay detection ranges were 2.8 to 9.5 log 10 50% cell culture infectious doses (CCID 50 )/g of liver or spleen and 1.8 to 8.5 log 10 CCID 50 /ml of serum. In samples presenting with undetectable liver or spleen virus, a value of ,2.8 was assigned (,1.8 for serum). Conversely, in cases wherein virus exceeded the detection range, a value of.9.5 (.8.5 for serum) was assigned. For statistical analysis, values of 2.8 or 9.5 log 10 (1.8 or 8.5 for serum) were assigned as needed for samples with undetectable or saturated virus levels, respectively. Detection of ALT in serum is an indirect method for evaluating liver disease. Serum ALT levels were measured using the ALT (SGPT) Reagent Set purchased from Pointe Scientific, Inc. (Lincoln Park, MI) per the manufacturer's recommendations. The reagent volumes were adjusted for analysis on 96-well microplates. Experimental design DEF201 dose range titration experiment. Hamsters were weighed on the morning prior to the day of infection and grouped (n = 15 for drug treatment groups, 26 for the placebo group) so that the average hamster weight per group across the entire experiment varied by less than 5 grams. Varying pfu amounts of DEF201, the rAd EV control virus, or saline placebo treatments were administered in a single i.n. dose 24 h prior to challenge with ,5 pfu of PICV. Five animals from each group were sacrificed on day 7 of infection. Serum was collected for assaying ALT activity, and virus titers were determined for liver, spleen, and serum samples as described above. The remaining 10 animals (21 for the placebo group) were observed 21 days for mortality and weighed individually every 3 days starting on day 0. Sham-infected normal controls (n = 3) were included for comparison. Extended pre-exposure prophylaxis experiment. The design was similar to the DEF201 titration experiment with the following differences. Hamsters were weighed on the morning of initial pretreatment (day 214 relative to the infection) and grouped (n = 15 per group). Groups were treated once i.n. with 10 8 pfu of DEF201, rAd EV control virus, or saline placebo. Treatments were given 14 or 7 days prior to challenge with ,5 PFU of PICV. Animals were observed for 28 days post-challenge for mortality. Post-exposure prophylaxis experiment. The design was similar to the DEF201 pre-exposure prophylaxis experiment with the following differences. Single dose i.n. treatments with 10 8 pfu of DEF201 or rAd EV were administered 24 h prior to, or 6, 24, or 48 h after challenge with ,5 pfu of PICV. On day 28 postinfection, the surviving animals (including 6 naïve sham-infected controls) were re-challenged. Morbidity and mortality were observed out to 58 days after the initial challenge. Kaplan-Meier survival plots and all statistical evaluations were done using Prism (GraphPad Software, CA). The log-rank test was used for survival analysis. For analyzing differences in viral titers, ALT levels, and weight change, a one-way analysis of variance (ANOVA) with Newman-Keuls post test or the Kruskal-Wallis (two-tailed) test with the Dunn's post test was performed based on Gaussian distribution of the data. In the initial trial, hamsters were treated with 10 6 to 10 8 pfu of DEF201 one day prior to challenge with a lethal dose of PICV. Pretreatment with the highest dose of 10 8 pfu of DEF201 resulted in 100% survival, and 10 7 and 10 6 pfu doses also significantly protected 90% and 60% of hamsters, respectively, from mortality ( Figure 1A) . Moreover, the hamster that succumbed in the 10 7 group, survived 19 days. Importantly, only one out of ten hamsters treated with 10 8 pfu of the control rAd EV virus survived the infection; however, there did appear to be a slight delay in the time to death in the hamsters that received the control virus treatment. The weights of the hamsters were measured every 3 days to assess weight gain over the course of the experiment as a marker of well being ( Figure 1B) . Notably, from day 3 to day 6, a time before weight loss due to illness from PICV infection would have been expected, hamster weights decreased as the dose of DEF201 increased. This would suggest that the higher treatment doses may have resulted in some loss of appetite, probably due to mild illness due to expression of consensus IFN since no overt effects were noticeable when handling the animals. The hamsters that received the 10 6 pfu dose of DEF201 gained weight through day 6 similarly to the animals treated with saline placebo and the normal controls (sham-infected, untreated) ( Figure 1B) . The high-dose of rAd EV control virus also resulted in a slight reduction in weight compared to the controls, suggesting that the immune response to the adenoviral vector alone may have caused some malaise in the animals. There was no elevation in serum ALT levels on day 7 of infection in samples collected from parallel treated and infected hamsters receiving DEF201 (Figure 2A ). Eighty percent of the rAd EV group and 100% of the PBS placebo group had elevated levels of ALT, reflective of liver disease. Interestingly, the 10 7 and 10 8 pfu DEF201 groups presented with little to no day-7 virus burden in the serum, liver, or spleen, while the 10 6 group developed viral titers that were comparable to the rAd EV and placebo controls ( Figure 2B-D) . A delay in the development of liver disease in the 10 6 pfu DEF201treated animals may explain the reduced ALT levels. Alternatively, saturation of liver virus titers in the low-dose DEF201, rAd EV, and placebo groups may have masked a substantial difference between the former and the viral vector and vehicle control groups. We next evaluated the prophylactic window of protection against PICV infection using the 10 8 pfu dose of DEF201. Animals were treated one or two weeks prior to challenge with a lethal dose of PICV. Consistent with the trend observed in initial dose titration study, hamsters treated with the 10 8 pfu dose of DEF201 had significantly reduced weights compared to those that received the rAd EV and placebo control treatments ( Figure 3A) . Nevertheless, the pretreatment with DEF201 seven days before infection was highly protective (90% survival rate; Figure 3B ). Notably, the single hamster that failed to survive the challenge succumbed on day 5, which was several days before the mean time to death measured in both the placebo and rAd EV groups. An autopsy to determine the cause of death was not performed. In hamsters treated two-weeks prior to PICV challenge, DEF201 significantly reduced mortality (50% survival) and extended the time of death in the animals that succumbed ( Figure 3C ). In contrast, uniform lethality was seen with animals that received the rAd EV and placebo treatments. Of the 5 surviving animals pre-treated with DEF201, one was anorexic at the conclusion of the study on day 28 post-infection. This was reflected by a 27% weight loss compared to the animals starting weight. It is possible that this hamster, which appeared ill and lethargic, was not able to completely prevent the infection. It was unclear whether it would have ultimately recovered if the observation period had been extended. On both the 7-day ( Figure 4A , C, E, G) and 14-day ( Figure 4B , D, F, H) pretreatments, DEF201 significantly reduced day-7 viral loads and liver disease (ALT) compared to the controls. The absence of elevated ALT levels in the DEF201-treated hamsters may be explained by the 2-3 log 10 reduction in liver virus burden ( Figure 4E , F) and a delay in the development of liver disease. Although tissue titers were slightly lower when DEF201 was given 7 days prior to challenge compared to the 14 day pretreatment, this was not evident with serum viral burden. Because most animals had measurable replicating PICV ( Figure 4C-H) , it is likely that survivors would have been immunized and protected from subsequent challenge. This may not be the case with hamsters treated with DEF201 24 h prior to challenge since most had no detectable virus titers in spleen, liver, or serum on day 7 of PICV infection ( Figure 2B-D) . Having observed dramatic protection when administered up to 2 weeks prior to challenge, a final experiment was conducted to determine the therapeutic value of DEF201 in the hamster PICV infection model. When DEF201 was administered 6 or 24 h after challenge, highly significant protection was observed ( Figure 5 ). Efficacy waned when DEF201 treatment was delayed to 48 h postinfection. As anticipated, the treatment given 24 pre-challenge verified previous activity, with all animals surviving challenge. Interestingly, there was higher than expected survival with the control rAd EV treatments initiated 24 and 48 h post-challenge, suggestive of a slight antiviral effect as the time of treatment was further delayed ( Figure 5) . The surviving hamsters from this experiment were re-challenged with PICV to assess the ability of DEF201 to enhance longer-term protection via acquired immunity. With the exception of 4 animals in the 24 h DEF201 pretreatment group, and a single animal in the rAd EV 48 h group, all animals that were challenged with PICV on day 0 of the experiment survived a second challenge Figure 3 . DEF201 extended pre-exposure prophylaxis protects hamsters from lethal PICV challenge. Animals were treated i.n. with a single dose 10 8 pfu of DEF201, the rAd EV control virus, or PBS placebo 7 or 14 days prior to PICV infection. Animal weights were measured two weeks prior to, and at the time of, PICV challenge. The effect of 7-day and 14-day pretreatments on A) weight change over the two-week period prior to PICV challenge and the extended PICV prophylaxis efficacy data for the B) 7-day and C) 14-day pretreatments are shown. ***P,0.001 compared to respective placebo-treated animals. b P,0.01, c P,0.001 compared to respective rAd EV-treated animals. doi:10.1371/journal.pone.0026072.g003 on day 28 ( Figure 5 ). All six naïve animals that were initially shaminfected succumbed as expected. In animals that were sacrificed on day 7 relative to the first infection, reductions in ALT and viral titers were most evident in the groups that received DEF201 within 24 h of infection ( Figure 6) . Notably, in the animals treated with the control rAd EV, there was an interesting trend that developed with the 6 h post-infection group having the greatest ALT levels and viral titers, followed by the 24, 48, and 224 h groups. This trend may suggest a low-level immune stimulation in the hamsters relative to the time at which the rAd EV was given. The resulting lack of measurable viral replication in the 224 h DEF201 group ( Figure 6B-D) is likely insufficient to elicit immunological memory. It is unclear as to why one of the first infection survivors from the 48 h rAd EV group ultimately succumbed to the second infection. In the present study, our findings demonstrate that expression of cIFN-a following a single i.n. administration of DEF201 offers a strong protective effect in hamsters against challenge with PICV that included limiting liver disease and inducing an antiviral state that inhibited systemic and tissue viral replication. The lack of significant antiviral activity elicited by the rAd EV control virus suggests that the enhanced antiviral response produced by DEF201 is largely due to the expression of the cIFN-a gene. The weak stimulatory effect seen in 1 of the 3 experiments was not surprising considering the number of host systems that play a role in sensing the adenovirus vector [24] ; however, the effect was short-lived. In contrast, the enhancement of the host antiviral defenses by DEF201 was long-lasting with a 14-day pre-PICV challenge prophylactic window. Moreover, DEF201 was effective when given 1-2 days post-PICV infection. These data also suggest that sufficient viral replication may be necessary to elicit an adaptive immune response that confers lasting protective immunity, as, for the most part, only re-challenged animals from the 24 h DEF201 pretreatment group succumbed to a second challenge with a lethal PICV inoculum. Presumably, the robust innate immunity and antiviral state induced by the DEF201 pretreatment rapidly controlled the ,5 pfu challenge dose obviating the development of the adaptive immune response and immunological memory. The pathogenic arenaviruses have evolved strategies to suppress and evade the host immune response [25, 26, 27, 28] , resulting in uncontrolled replication and broad dissemination. However, they appear to be unable to block the induction of IFN stimulated genes via exogenous type I IFN [29] , which may, in part, explain the success of DEF201 and cIFN-a treatments [15] . Also essential to the success of DEF201 was early intervention prior to significant viral replication and engagement of innate immune suppressive functions. Indeed, early induction of a strong type I IFN response is associated with favorable disease outcome in nonhuman primates challenged with Lassa virus [30] . Early post-exposure prophylaxis was also required with exogenous cIFN-a protein administered by the i.p. route [15, 16] . With the multiple strategies that arena and other pathogenic viruses have in place to subdue the IFN-mediated host antiviral response [31] , the utility of DEF201, recombinant IFN proteins, and IFN inducing agents will depend upon the nature of the IFN pathway blockade and require early administration to be effective post-exposure. Notably, with daily cIFN-a protein injections of up to 40 mg/kg, significant protection was observed; however, survival rates did not exceed 80% in those studies employing the same PICV hamster model system and virus stock [15, 16] . In contrast, DEF201 consistently elicited greater protection (90-100%). The improved efficacy observed with DEF201 may be explained by a combination of factors that includes constitutive expression of fully glycosylated protein and reduced animal stress levels by avoiding daily injections for 7-10 days. We hypothesize that with the appropriate dose of DEF201, therapeutic levels of consensus IFN-a can be maintained, effectively eliminating the daily bolus effect produced by i.p. injections. In addition, because cIFN-a is produced in genetically engineered Escherichia coli, the native glycosylation pattern is lost. Conceivably, enhanced immunotherapeutic activity results from fully glycosylated cIFN-a expressed from cells transduced with DEF201. Previous studies in mice with a related DEF201 virus expressing mouse IFN-a (mDEF201) have shown the utility of adenovirusbased system to counter viral infections [20, 32, 33] . More recently, in a different hamster model of viral hemorrhagic fever, several of us reported on efficacy of DEF201 in mitigating YFV infection and disease [19] . YFV infection appears to be more sensitive to the effects of DEF201, as a lower dose was able to provide complete protection. Taken together with the results of the present study, the experimental animal data support the broad use of DEF201 for extended pre-exposure and early post-exposure prophylaxis applications. Further investigations using advanced arenavirus models based on challenge of nonhuman primates with pathogenic arenaviruses [17] are needed to better evaluate the potential of DEF201 to prevent severe disease in humans. Nonhuman primate models should allow the full spectrum of cIFN-a activity not possible in hamsters or guinea pigs. The familiarity of the FDA with adenovirus gene delivery technology and approved cIFN-a protein support the development of DEF201 for clinical use. An important step in the development process is the safety/toxicology testing in rodents, which is presently underway. In our studies, the highest dose of 10 8 pfu of DEF201 administered by the i.n. route appeared to be well-tolerated in hamsters despite evidence of weight loss. They did not appear visibly ill, but clearly the treatment was having some effect that possibly led to reduced food and water consumption consistent with mild toxicity seen with IFN-a therapy. The i.n. delivery route is designed to circumvent pre-existing immunity to adenovirus type 5 in humans [21, 34] , and may limit systemic inflammation that could occur by parenteral administration of large numbers of adenovirus particles. Ultimately, the production of a shelf-stable, powdered formulation of DEF201 for easy i.n. administration and long-term storage would be ideal for stock-piling in the event of the need for mass distribution due to intentional release or (re)emerging disease outbreaks of arena or other viral etiology.

Protein Disulfide Isomerase and Host-Pathogen Interaction

Reactive oxygen species (ROS) production by immunological cells is known to cause damage to pathogens. Increasing evidence accumulated in the last decade has shown, however, that ROS (and redox signals) functionally regulate different cellular pathways in the host-pathogen interaction. These especially affect (i) pathogen entry through protein redox switches and redox modification (i.e., intra- and interdisulfide and cysteine oxidation) and (ii) phagocytic ROS production via Nox family NADPH oxidase enzyme and the control of phagolysosome function with key implications for antigen processing. The protein disulfide isomerase (PDI) family of redox chaperones is closely involved in both processes and is also implicated in protein unfolding and trafficking across the endoplasmic reticulum (ER) and towards the cytosol, a thiol-based redox locus for antigen processing. Here, we summarise examples of the cellular association of host PDI with different pathogens and explore the possible roles of pathogen PDIs in infection. A better understanding of these complex regulatory steps will provide insightful information on the redox role and coevolutional biological process, and assist the development of more specific therapeutic strategies in pathogen-mediated infections.

Host cells have the ability to cope with the progression and severity of infection in response to different types of pathogen. On the other hand, numerous mechanisms have evolved that support the use of the host cell machinery to facilitate pathogen survival and multiplication. Such co-evolutionary processes are directly affected by different physicochemical factors within different cell compartments, both in the host and in pathogen. For instance, pH critically affects antigen stability of the influenza virus which modulates endosome acidity that attenuates its own infection [1] . ROS (and Reactive Nitrogen Species) production and the redox state of different cell compartments are also critically involved in cellular hostparasite interaction. Among the many redox sensitive proteins that are altered during the course of different infections, protein disulfide isomerase (PDI-) mediated redox switches have been associated with pathogen attachment-internalization, antigen processing in the ER/phagosome, and the regulation of ROS production by Nox family enzymes. Thus, PDI emerges as a ubiquitous redox protein that regulates different steps of diverse infection processes. Several pathogens also have their own PDI that act as an important virulence factor (Table 1 ). Other redox modifications directly mediated by ROS and especially via nitric oxide (NO) generated by inducible nitric oxide synthase (iNOS), which is abundant in phagocytic cells, have been reviewed elsewhere [2, 3] and are not considered in this article. Below, the main cellular redox aspects of host and pathogen PDI will be discussed. The ancient PDI is a ubiquitous redox chaperone belonging to the thioredoxin oxireductase super family and can reduce (reaction 1), oxidize (reaction 2), and catalyse dithiol-disulfide exchange reactions (i.e., isomerase activities, reaction 3, Figure 1 ). Such broad range of activities overlaps with the chaperone role of PDI that overall performs a housekeeping function in helping to maintain proteins in a more stable conformation. There are around 20 PDI homologues, and the detailed structure and function of eukaryotic PDIs have been covered in recent excellent reviews [4, 5] . The classic mammalian PDI (55 kDa) has several domains ordered as a-b-b -a -c with 2 thioredoxin-like motifs (Trp-Cys-Gly-His-Cys) displayed in the a and a domain [4] [5] [6] (Figure 1 ). PDI is abundant in the ER (∼ 0.5 mM) where the relatively oxidizing conditions at basal level (i.e., GSH/GSSG ratios ∼ 2-3 : 1) favours PDI isomerase/oxidase activity, which is primarily involved in client protein redox folding (reaction 2-3, Figure 1 ). The oxidizing equivalents for this process are driven mainly by the ER thiol-containing oxidase, Ero1 (endoplasmic reticulum oxidase-1), which binds FAD and is in turn re-oxidized via electron transfer to oxygen, generating H 2 O 2 in the process [7] [8] [9] [10] . The H 2 O 2 destiny is elusive, but it can oxidize ER-located peroxiredoxin IV (PrxIV) that is further reduced by PDI that is oxidized in the process [11] . This redox circuit is thought to increase total protein folding and thiol oxidation via Ero1 [11] . However, even in the absence of Ero1, protein folding still occurs, and it is suggested that other oxidases may compensate for redox demand in the ER in some circumstances [12, 13] . Nevertheless, the PDI-Ero1-dependent oxidative activity is balanced to cytosolic glutathione levels suggesting a functional redox interplay between these compartments [12] . PDI reductase activity has been primarily associated to more reducing compartments (i.e., GSH/GSSG ratios ∼ 30-100 : 1), such as those in the vicinity of the plasma membrane [6, 10] . PDI redox versatility is mainly governed by the low pKa of the proximal cysteine on the active N-terminal a domain. Indeed, the lower pKa of 4.5 renders PDI a much better oxidase than thioredoxin, which has a pKa of 7.1 and is mainly a reductase in most neutral pH cell compartments. It should also be noted that PDI functions as a chaperone independently of its redox-active domains as especially required for its ATPase and Ca 2+ activity, although PDI redox motifs still stabilize binding interaction [4, 5, 10] . In the ER, PDI is tightly associated with prolyl-4 hydroxylase (the rate-limiting enzyme for collagen biosynthesis), the Sec61 translocon, and the MHC class I complex (see later). It can be also found as a heterodimer with microsomal triglyceride transfer protein [6, 10] . PDI is a soluble homodimer and does not have a transmembrane domain and, similarly to other ER chaperones, carries the KDEL C-terminal sequence which binds to respective receptors in the COP vesicles that circulate in the ER-Golgi vicinity, recycling proteins back to ER. PDI also undergoes intense intracellular trafficking and is found on the surface of diverse prokaryotic and eukaryotic cells [14] [15] [16] . Despite limited knowledge about this traffic, it is possible that PDI exits the ER through the translocon Sec61 pore and/or via secretory vesicles [17] . PDI is thought to attach to lipids, glycans, and integral membrane proteins via electrostatic interactions at the cell plasma membrane [14, 15] , where its reductive activity mediates the infection of different pathogens ( Figure 2 , discussed later). PDI along with ERp57 have been found in the nuclei in association to DNA and affecting transcriptional activity of NF-kb, AP-1, and STAT3 [15] . These transcriptional regulators are key elements in many inflammatory processes, but their functional association to PDI and to different pathogen still elusive. In contrast to many other members of its family, such as thioredoxin itself and Erp57, PDI is not normally found in the cytosol, where it is likely cleaved by caspase-3 and -7 [18] . Protozoans and bacteria have their own PDIs ( Table 1 ). The function of these PDIs in protein folding is poorly understood; however, there are intriguing data correlating PDI expression and the pathogenicity of several parasites, especially obligatory intracellular protozoans. Leishmania that leads to distinct types [20] . If L. chagasi is a subgroup of L. infantum brought to America by European colonist or other specie in its own still a matter of controversy (see discussion in [21] ). The infection cycle in the vertebrate host and is initiated when Leishmania promastigote is injected into the skin by the insect vector. In the host, the promastigote is phagocytised especially by macrophages, and further it is converted into intracellular amastigote. Amastigote replicates inside the phagosome within the cell and is liberated after the cell lyses, subsequently infecting other cells resulting in the progression to disease [21, 22] . L. amazonensis has at least four PDIs, and the use of specific PDI inhibitors substantially affected parasite growth [23] . In L. major, the increased levels of Leishmania PDI (LmPDI) expression and secretion at the parasite surface reflects optimal protein folding balanced to parasite multiplication. Importantly, this is correlated to high virulence of the parasite strains [16] . More recently, the use of LmPDI antigens to generate a vaccine for L. major partially protected BALB/c animals and accelerated the cure of different strains of mice [24] . Similarly to Leishmania species, other parasites of the trypanosomatid group such as Trypanosoma contain several genes predicted to encode for PDIs, that can execute N-glycosylation and protein folding in the ER [25, 26] . Although PDI was considered essential for T. brucei survival, PDI activity was not essential for the growth of trypanosomes in vitro [26] . PDI is also expressed in different species of Plasmodium protozoans (Table 1) , the parasites that cause malaria [27, 28] . P. falciparum expresses at least nine different PDIs and the PfPDI-8 has great similarity to the prototype PDI and is expressed during all stages of parasite life cycle. This PDI facilitates the disulfide-dependent conformational folding of EBA-175 protein, an emerging candidate for the development of malaria vaccines [28] . This is intriguing given that malaria parasites express proteins with high content of cysteine, which are associated to parasite invasion and sequestration in the vertebrate host and transmission into mosquito host [28] . Finally, Toxoplasma gondii PDI was identified in host tears, suggesting an extracellular location and adhesion to host cells during the initial phase of infection [29] . Antigen presentation occur through two distinct pathways. Antigen presenting cells (APCs; especially macrophages and dendritic cells; DCs) are long-lived cells that capture antigens and subsequently process and present them at the cell surface, where they are recognized by T-lymphocytes. This process provides a long-term adaptive immune response to fungi, bacteria, and parasite. After internalization by the APC, antigens pass through phagosome/lysosome vesicles, where they form complexes with MHC class II (Figure 2 ), which are recognized by helper CD4+ T lymphocytes (exogenous pathway). In contrast, self cell antigens and virus synthesized within cells (mostly non-APCs) are degraded by the proteasome in the cytosol and nucleus. In successive steps, the antigen is processed, folded, and incorporated into the MHC class I ( Figure 2 ) and the complex exposed on the cell surface, and recognized by cytotoxic CD8+ T lymphocytes (endogenous pathway). These two pathways overlap and some antigens are presented by both MHC class I and II, in a process called cross-presentation. This has been described in DCs responding to viral infection, transplant rejection, and some autoimmune diseases and cancer. Moreover, a wide range of pathogens passing or living in the phagosome such as Mycobacterium tuberculosis, Salmonella typhimurium, Toxoplasma gondii, and especially Leishmania spp and Trypanosoma cruzi are all crosspresented in association to high levels of CD8 + T cells [31] . PDI as part of the ER protein folding machinery directly regulates antigen processing of the MHC class I complex [32] [33] [34] [35] . Antigens that are degraded by peptidases and proteasome to shorter peptides in the cytosol and nucleus can be further transported to the ER through the TAP system, a transmembrane ER type of ATP-binding cassette (ABC) peptide transporter family [36] . ER-located PDI interacts with the peptide-loading complex (PCL) that efficiently promotes peptide assembly with MHC class I molecules and supporting the exit of the peptide-antigen complex from the ER [32] [33] [34] . Other PCL components include calreticulin, tapasin, ERp57 (another PDI family member), and the TAP transporter itself. Cells lacking PDI present much less peptide loading to MHC class I and the disulfide bridge between the peptide and MHC groove remains in a reduced redox state [32] . Normally, this interaction is affected by the redox exchange between PDI (predominantly oxidized) and ERp57 (predominantly reduced) [32] , a condition in which PDI favours the release of peptide-MHC class I from the PCL and the antigen-MHCI complex is exited from ER [34] . In fact, PDI-bound peptide facilitates the disassembly of the tapasin-ERp57 complex while the PDI unbound to the complex is unable to interact with tapasin-ERp57, retaining MHC I molecules in the ER [34] . Overall, PDI redox activity modulates the stability of the antigen peptide-MHC class I complex and further determines the transport of the complex to the plasma membrane [32] [33] [34] [35] . These redox effects may vary according to the type of antigen and some pathogens interfere with this pathway to escape antigen process and evading CD8 + T-cells recognition. This is the case for US3 protein from human cytomegalovirus, which enhances PDI degradation via the proteasome [32] . PDI participation in immune response, however, goes beyond its role in the ER protein folding machinery and it acts at other cellular steps of host-pathogen interaction. PDI in the ER is also thought to play a role in parasite phagocytosis, and the PDI displayed on the cell surface can mediate the entry of some viral, bacterial, and protozoan. PDI is also implicated in protein unfolding and trafficking of some pathogenic antigens across the endoplasmic reticulum and towards the cytosol by the endoplasmic reticulum-associated degradation system (ERAD). This is the main pathway where proteins are retrotranslocated from the ER to cytosol and further degradated by the proteasome. Next, we discuss some examples of the cellular association between host PDI and different pathogens. Phagocytosis is the main gate for large microbes to enter into APCs. After binding and attaching to the pathogen, these cells can internalize organisms and large particles even bigger then their own size, which are then phagocytosed in an active process that involves intense membrane remodelling [37] . Proteomics studies accumulated over the last decade revealed the presence of ER chaperones in the isolated phagosome, uncovering a process called ER-mediated phagocytosis [38] [39] [40] [41] [42] [43] [44] [45] . ER chaperones were detected in phagosomes of macrophages exposed to different particulate material and pathogens, including latexbeads opsonised or not with immunoglobulin G (IgG) or mouse serum (to facilitate entry through the FcR or complement receptors), IgG-opsonized erythrocytes, promastigotes of Leishmania donovani derived from wild-type cells or cell-surface LPG knockout, among other parasites [38] . A mix of ER and endocytic vesicles in the formation of the parasitophorous vacuoles (PVs) during the uptake of different Leishmania spp. was recently shown in macrophages overexpressing ER-tagged green fluorescent protein [41] . The presence of the ER proteins Sec61, Bip/Grp78, and PDI in the phagosome of APCs [38, 41] support the idea that the ER provides the necessary machinery for antigen translocation from the phagosome to the cytoplasm and thus, possibly converges MHC class I and class II antigen cross-presentation [42] (Figure 2 ). There are several other complementary hypotheses on how peptides cross from phagosomes to cytoplasm during the cross-presentation process [43, 44] . Neutrophils are short-lived cells (half-life of 4-8 h in the human circulation) and very active in the phagocytosis of large microbes such as bacteria, parasites, and fungus. Contrary to APCs, neutrophils only contain restricted amounts of ER machinery and are thought to lack the ER-mediated phagocytosis process [38] . Whether ER proteins functionally operate on phagocytosis-mediated infection has not been well characterised yet. An important work has shown that Dictyostelium lacking both ER calreticulin and calnexin present altered phagocytic cup formation and substantial decline in phagocytosis [45] . These two proteins utilise Ca, and their disruption per se affects actin filaments and plasma membrane remodelling during phagocytosis [45] . We recently showed that PDI is critically involved in Leishmania parasite infection in vitro [22] . We showed that phagocytosis of promastigotes (but not amastigotes) of Leishmania chagasi was significantly inhibited by macrophage incubation with the thiol/PDI inhibitors DTNB, bacitracin, phenylarsine oxide, and neutralizing PDI antibody in a parasite redox-dependent way [22] . The phenylarsine response is of particular interest, since this arsenic compound may act similarly to antimonials, widely used in leishmaniasis chemotherapy [46, 47] . PDI preferentially affects parasite internalization and the phagocytosis of the promastigote forms is increased when wild-type PDI is overexpressed in macrophages, an effect opposed by PDI knockdown. At later stages of infection (i.e., after 4 h), PDI from promastigote-infected J774 macrophages was immunoprecipitated and subsequently blotted with an anti-Leishmania antibody revealing a parasite band at ∼ 94 KDa [16, Figure 10 (b), lane 5]. Subsequent removal and analysis of this band by mass fingerprint spectrometry showed a 58% match with elongation factor 2 (EF2) of L. major (Q4Q259; data not shown). The incubation of purified bovine PDI (Sigma, P3818) and parasites did not yield any detectable protein complexes, suggesting that the macrophage milieu may be important to sustain PDI-EF2 association [22] . Interestingly, Leishmania EF2 has important virulent features and acts as a soluble antigen in lymphocyte stimulation in vitro [48] and in vivo [49] . Moreover, proteomics studies revealed that EF2 is secreted during promastigote differentiation into the amastigote stage with potential immunomodulatory proprieties in animal models [50] . Leishmania EF2 is therefore of particular interest for Leishmania therapeutic interventions such as vaccines. Although our studies did not address the role of the ER in mediating phagocytosis, these data provide compelling evidence for a functional role of ER-PDI in a host-parasite interaction. Other mechanisms underlining PDI-mediated L. chagasi promastigote phagocytosis involves its association to ROS production by phagocyte NADPH oxidase and this is discussed next. The NADPH oxidase (Nox) family of enzymes uses NADPH as an electron donor to convert oxygen to superoxide anion (O 2 •− ), a precursor of H 2 O 2 and other powerful oxidants such as hydroxyl radical and peroxynitrite (in the presence of nitric oxide), collectively called ROS [3, 51, 52] . Each of the seven oxidase family members is characterized by a distinct catalytic subunit (i.e., Nox1-5 and Duox1-2), and has differing requirements for additional protein subunits [51, 52] . The prototypic member of the Nox family, Nox2 oxidase (or gp91 phox oxidase), is best known for its role in neutrophil and macrophage phagocytosis. Genetic defects in the enzyme are related to chronic granulomatous disease, a condition in which affected children suffer from recurrent severe fungal and bacterial infections due to defective phagocyte function [51] . Each Nox isoform forms heterodimers with a lower molecular weight p22 phox subunit and is predicted to be membrane-bound. Nox2 is normally quiescent and acutely activated by agonists such as PMA, LPG, and cytokines in a tightly regulated process in which cytosolic subunits (p47 phox , p67 phox , p40 phox , and Rac1 in the case of macrophages and dendritic cells, or Rac2 in neutrophil) associate with the Nox2-p22 phox heterodimer to initiate enzyme activity [51, 52] . Nox2 also has electrogenic features [53] and in APC cells is linked to the regulation of phagosome/lysosome pH and antigen processing [54, 55] . Usually, phagosome acidity is maintained by a vacuolar ATPase (V-ATPase) that transports protons from the cytosol into the phagosome lumen, therefore regulating the function of lysosome proteases in the fused phagolysosomes. Savina et al. [56] [57] [58] have shown that Nox2-derived superoxide in the phagosomal vesicle promptly consumes protons maintaining a higher pH ambient in dendritic cells during particle internalization, which favours antigen processing and presentation [56] [57] [58] . Opposite results were found in macrophages [56] [57] [58] . The selective role of Nox2 in different phagocytic cells remains to be defined. The jury is out on whether the results shown in macrophages (association of Nox2 to Rab27a; a member of Rab family of GTPases) are related to vesicle traffic molecule assembly and quality, or rather associated to degradation processes [59] . Nox complex protein expression and function is greatly affected by redox compounds, and it is especially regulated by PDI with implications for cell signalling [60] . The association of PDI to p22 phox and other Nox isoforms in different cell types, especially in vascular cells, has been previously described [60] [61] [62] . A functional and spatial/physical interaction between PDI and the p22 phox oxidase subunit was shown in macrophages [22] and more recently between PDI and p47 phox in neutrophils [62] . In macrophages, PDI-Nox association was correlated to Leishmania infection in vitro [22] . It is well known that during phagocytosis of Leishmania, Nox2 is activated and parasite uptake is inhibited by antioxidants such as catalase [22] . Intriguingly, in the course of promastigote infection, some parasites evade that stressful condition and convert themselves into intracellular amastigotes, multiplying and resulting in progression to a disease process. Overall, our studies support the view that parasite phagocytosis/infection by macrophages is a redox process mediated by PDI in at least two ways. Initially, PDI-NADPH oxidase increases ROS production generating an oxidizing milieu, which seems to favour promastigote infection. The downstream role of ROS generated by PDI-NADPH oxidase remains unknown but can be related to the unfolded protein response signalling [63] or, similar to PDI-Ero1, to protein folding in the macrophage ER compartment with key implications for antigen processing. Nevertheless, at later stages of infection, macrophage PDI physically associates with Leishmania elongation factor-2 (as discussed earlier). Some viruses envelop their genetic material within a protein-coated capsid in a further lipid membrane layout, for example, influenza virus, baculovirus, hepatitis-C, HIV, and Herpes virus. These enveloped particles require successive steps to successfully entry and infect host cells. They usually first attach onto host receptors (and attachment factors), and their membranes fuse to interact with endosome vesicles that traffic the virus toward the endoplasmic reticulum, where it is uncoated. The proteins are finally transported to the cytosol and nucleus [64] (Figure 2 ). There is convincing evidence showing that most viral infections are strongly influenced by changes in the redox environment and that host PDI mediates infection of enveloped viruses [65] [66] [67] [68] [69] [70] . In the course of HIV infection, the virus first binds to attachment factors, for example, mannose binding C-type lectin receptor and intracellular adhesion molecule (ICAM-3) on the surface of host CD4 + T cells. The glycoprotein 120 (gp120) subunit of the virus envelope binds to immunoglobulin G of CD4 + and undergoes conformational changes, allowing the virus to interact with its coreceptors, CXCR4 or CCR5. These interactions favour downstream conversions of gp41 envelope subunit to a competent fusion conformation. Initial studies showed that membrane-impermeable PDI inhibitors and monoclonal antibodies against PDI prevent HIV-1 infection [65] . It was then revealed that the domain D2 of the CD4 has redox-active disulfide bonds and is regulated by thioredoxin [66] . Using membrane-impermeable reducing agents (especially arsenical-derived compounds) and labelling thiol reagents, it was demonstrated that CD4 + reactive thiols critically drive HIV entry into cells [66] . Work from another group also revealed that PDI, on the surface of HIV-1 target cells, reduces disulfide bonds of the recombinant envelope glycoprotein gp120 (reaction 1, Figure 1 ), a reaction prevented by the usual PDI inhibitors [67] . Intriguingly, PDI silencing in U373 and HeLa cells had little impact on HIV infection itself as compared to the effect mediated by general thiol inhibitors [68] . The reasons for this discrepancy remain to be elucidated and raise the question whether the reductive effect of PDI is coupled to other redox proteins (e.g., thioredoxin or Nox's) that could amplify virus-CD4 redox association in some cells. It is noteworthy that in these later studies, PDI knockdown on the cell surface was not evident as compared to massive loss of most PDIs within the ER; an observation that supports the idea that PDI in the ER has little impact in HIV-mediated infection [68] . Thiol inhibitors also affect viral fusion as that mediated by the fusion (F) protein from the Paramyxovirus Newcastle disease virus [69] . The overexpression of PDI family members PFDI and ERdj5 has also been shown to significantly catalyze the reduction of thiols in F protein, facilitating membrane fusion [70] . There is evidence suggesting a possible association between PDI and infection mediated by some members of the of Herpesviridae viruses family [71] . PDI is also implicated in the attachment of some bacteria from different species of Chlamydia [72] [73] [74] . Chlamydia is an obligatory intracellular pathogen that causes diverse diseases in humans. The most common species are Chlamydia trachomatis, which is sexually transmitted and can cause blindness and infertility, and C. pneumoniae, which affects the respiratory tract. CHO cells have impaired endogenous PDI expression due to a defect in truncated mRNA processing, thus providing a valuable model to understand the effect of PDI-mediated cell-cell interaction and infection. These cells are very resistant to Chlamydia infection showing impaired attachment, an effect restored by ectopic expression of PDI [73] . Similar to HIV infection, the molecular mechanisms most likely include the reductive activity of PDI (reaction 1, Figure 1 ) on the surface of CHO cells [72] . Crossing the endoplasmic reticulum (ER) membrane is an irreversible process for most proteins. In some cases, however, this flow is reversed and misfolded proteins retained in the ER are retrotranslocated to the cytosol via ERAD to be degraded by the proteasome. This pathway is also exploited by small pathogens, especially non-enveloped viruses and some bacterial toxins, to gain access to the cytosol. In these cases, antigenic particles that reach the ER by different means suffer molecular redox rearrangements and binding to PDI allowing them to be transported back to the cytosol or nucleus. Well-known examples are infections mediated by Polyomaviruses (Py) and Simian virus 40 (SV40) extensively studied in the field of carcinogenesis. After SV40 interaction with the GM1 receptor on the cell surface, the particle enters the host cell through endocytosis and traffics via the caveosome (a particular caveolin containing endosome with neutral pH) towards the ER compartment [75, 76] . SV40-coated pentamers are linked to each other by disulfide bonds between cysteine 104 (C104). Further isomerisation in the ER is crucial for the viral uncoating process. In vitro cell screening shows that among all ER-resident proteins, PDI and ERp57 more specifically regulate SV40 infection [75] . PDI silencing substantially decreases SV40 infection that is also dependent on some redox sensitive cysteines on the viral particle [75] . PDI cooperation with ERassociated ERAD proteins Derlin-1 and Sel1L is Ca dependent and facilitates SV40 traffic through ERAD [75] . A similar pathway is used by some nonobligatory intracellular bacteria that exert their effect through production of potent endotoxins, such as diphtheria toxin (DT) and cholera toxin (CT). These proteins function similarly to some plant toxins, such as ricin and abrin. Conversion into toxic proteins involves cleavage of their interchain disulfide bond, allowing them to traffic into the endocytic pathway within the host cell [77, 78] . In humans, CT is derived from the Bacterium Vibrio cholerae that causes cholera disease and has 2 subunits (A1 and A2). The protein first attaches to the host cell surface via GM1 and the subunit A2, which contains a KDEL sequence, and is transported back to the ER (see earlier discussion). There, PDI reduces and unfolds A2 and A1 that exit the ER via the Sec61 channel into the cytosol. PDI in the reduced state (reaction 1, Figure 1 ) binds to the toxin and subsequent oxidation of PDI, probably via Ero1α, enables the release of CT toxin [79, 80] . The active polypeptide A1 efficiently modifies a heterotrimeric G protein in the cytosol that leads to massive loss of chlorine and water secretion by intestinal epithelial cells in mammals, resulting in severe diarrhoea. In this article we have reviewed the main cellular aspects of PDI-mediated host pathogen interactions and the pathways that are involved in viral, bacterial (including bacterial toxins), and parasitic infections. A number of cellular mechanisms through which PDI modulates some specific cellular pathways in immune cells have been described, such as redox-sensitive attachment, antigen presentation in the ER and exit from it, and association to phagosome and ROS production by NADPH oxidase (Figure 2 ). Many of these responses are antigen-specific and the precise mechanisms of action remain to be fully elucidated, especially in the context of redox changes in cross-presentation phenomena. Moreover, little is known about the role of PDI in infection per se, as well as how PDI signals to a more integrated cellular response to stress [63] . PDI global knockout mice are only viable until birth, but partial gene-modified mice and also modified pathogens will help to reveal the significant redox role of PDI and its redox partners. Overall, PDI is a key regulator that may propagate or limit the severity of the infection processes, depending on the infectious organism involved. A better understanding of these complex regulatory steps will provide insightful information on the redox role and coevolutional biological process, and assist the development of more specific therapeutic strategies in pathogen-mediated infections. Endoplasmic reticulum MHC: Major histocompatibility complex H 2 O 2 : Hydrogen peroxide ERp57: A member of PDI family also know as glucose-regulated protein or 58-kD (GRP58) NF-kB: Factor nuclear kappa B AP-1: Activator protein 1 STAT-3: Signal transducer and activator of transcription 3.

Improved Immunological Tolerance Following Combination Therapy with CTLA-4/Ig and AAV-Mediated PD-L1/2 Muscle Gene Transfer

Initially thought as being non-immunogenic, recombinant AAVs have emerged as efficient vector candidates for treating monogenic diseases. It is now clear however that they induce potent immune responses against transgene products which can lead to destruction of transduced cells. Therefore, developing strategies to circumvent these immune responses and facilitate long-term expression of transgenic therapeutic proteins is a main challenge in gene therapy. We evaluated herein a strategy to inhibit the undesirable immune activation that follows muscle gene transfer by administration of CTLA-4/Ig to block the costimulatory signals required early during immune priming and by using gene transfer of PD-1 ligands to inhibit T cell functions at the tissue sites. We provide the proof of principle that this combination immunoregulatory therapy targeting two non-redundant checkpoints of the immune response, i.e., priming and effector functions, can improve persistence of transduced cells in experimental settings where cytotoxic T cells escape initial blockade. Therefore, CTLA-4/Ig plus PD-L1/2 combination therapy represents a candidate approach to circumvent the bottleneck of immune responses directed toward transgene products.

Since the original reports describing the use of adeno-associated virus (AAV) vectors for transfer of β-galactosidase gene to muscle (Kessler et al., 1996; Xiao et al., 1996; Fisher et al., 1997) , recombinant AAVs (rAAV) have emerged as very efficient and potentially non-immunogenic vector candidates for delivering therapeutic genes to a variety of tissues and treating monogenic diseases. As they poorly activate innate immunity and weakly transduce dendritic cells, rAAV appear as far less immunogenic than adenoviral vectors (Zhang et al., 2000; Zaiss et al., 2002; McCaffrey et al., 2008) . Nevertheless, it has rapidly become clear that rAAV vectors carrying various transgenes can, under different conditions, induce potent immune responses that could ultimately lead to destruction of transduced cells in vivo and consecutive disappearance of transgene expression (Manning et al., 1997 (Manning et al., , 1998 Halbert et al., 1998; Brockstedt et al., 1999) . Not surprisingly therefore, their use has even been proposed in genetic vaccination protocols aimed at eliciting cellular and humoral immune responses against different microorganisms (Kuck et al., 2006; Du et al., 2008) . In primates and humans, rAAV administration has also been documented to elicit significant cytotoxic CD8 + T cell responses directed against the viral as well as the transgenic "exogenous" proteins, resulting in the destruction of transduced cells and complete loss of transgene expression (Manno et al., 2006; Mingozzi et al., 2007; Gao et al., 2009) . Additionally, on the side of the humoral immunity, production of neutralizing antibodies targeting capsid proteins may also prevent vector readministration and accelerate the loss of the therapeutic protein through the formation of immune complexes. Such immune complexes may further sensitize the cellular immune response by enhancing cross-presentation of the transgenic protein by the antigen-presenting cells (APC). Therefore, developing strategies to circumvent immune responses and facilitate long-term expression of transgenic therapeutic proteins has been identified as one of today's main challenges for the translation of rAAV vectors into the clinic (Mingozzi and High, 2011a,b; Nayak and Herzog, 2011) . Depending on the experimental situation, rAAV-mediated gene transfer can either lead to durable transgene expression or, conversely, to the rapid formation of neutralizing antibodies and/or destruction of transduced cells by cytotoxic cells. Several factors influencing the immune response against transgenic proteins encoded by the rAAV vectors have now been identified including host species, route of administration, vector dose, immunogenicity of the transgenic protein, inflammatory status of the host and capsid serotype (Mays and Wilson, 2011) . These factors are thought to influence immunogenicity by triggering innate immunity, cytokine production, APC maturation, antigen presentation and, ultimately, priming of naïve T lymphocytes to functional effectors. Therefore, the idea to dampen immune activation by interfering with these very mechanisms has logically emerged with the aim to induce a short-term immunosuppression, avoid the early immune priming that follows vector administration and promote long-term tolerance (Zaiss and Muruve, 2008) . Here, we evaluated two different strategies to inhibit the undesirable immune activation that follows muscle gene transfer by acting at two different checkpoints of the immune response, i.e., on T cell priming or on the functions of activated T cells that may escape such priming blockade. We used the administration of CTLA-4/Ig to inhibit the substantial immune priming that immediately follow vector injection. Indeed, CTLA-4/Ig represent a potent immunosuppressive fusion protein that reversibly prevents T cell activation (Wallace et al., 1995) and is now used in the clinic to treat inflammatory diseases such as rheumatoid arthritis (Bluestone et al., 2006) . Its immunomodulatory action depends on its competitive inhibitory effect on the CD28/B7 pathway thereby preventing the pivotal CD28-dependent costimulation required to fully activate T lymphocytes (Salomon and Bluestone, 2001) . As a second strategy, we turned to immunomodulatory molecules that could protect transduced muscle fibers from immune attacks by activated T cells. For that, we aimed at stimulating the inhibitory PD-1 molecule expressed on T cells upon activation, by the gene transfer of its ligands PD-L1 or PD-L2 to muscle cells (Freeman et al., 2000; Latchman et al., 2001; Ishida et al., 2002) . We show herein that acting on these two non-redundant mechanisms of tolerance provides synergistic effects that prolong transgene expression in muscle even in the presence of circulating cytotoxic T cells directed against the transgene product. Female C57BL/6 (B6) mice were obtained from Centre d'Elevage Janvier (Le Genest Saint Isle, France). Mice were all between 7 and 10 weeks of age at beginning of experiments. For transduction with rAAV vectors, mice back legs were shaved under general anesthesia and titrated 1 × 10 11 vector genomes (vg) rAAV2/1-Ova or rAAV2/8-Ova were injected (50 μl in PBS) in the gastrocnemius muscles. Where indicated, 10 11 vg rAAV2/1-PD-L1 or rAAV2/1-PD-L2 were mixed with 10 11 vg rAAV-Ova and co-injected using the same procedure. For costimulation blockade experiments, mice were injected i.p. with 200 μg CTLA-4/Ig (Chimerigen laboratories, MF-110A4) diluted in 200 μl of PBS. In some experiments, lymphocytes from tolerized mice were transferred to conditioned recipients to further evaluate the presence of anti-Ova lymphocytes and to study their functionality in vivo. For that, 50 × 10 6 splenocytes harvested from individual mice that have be treated 80 days before with AAV-Ova, with or without immunomodulatory regimens, were injected i.v. into 5 Gy-irradiated syngenic C57BL/6 recipient. One day after, each individual mouse was shaved and injected subcutaneously with 1 × 10 6 syngenic EG7 tumor cells expressing the Ova antigen and known to be sensitive in vivo to CD8 + T cell cytotoxicity in primed animals (Moore et al., 1988) . Tumor sizes were measured with a digital caliper three times a week during 26 days. Tumor volume was calculated as length × width × [(length + width)/2]. All animal experiments were approved by the local institutional ethic committee for animal experimentation (authorization #0211-22 "Comité Régional d'Éthique en Expérimentation Animale de Normandie"). The rAAV-Ova vector, a kind gift of Roland W. Herzog, was described previously (Wang et al., 2005) . The cDNA encoding mouse PD-L1 (CD274) or PD-L2 (CD273) were cloned using standard molecular biology procedures and introduced in the SSV9-CAG plasmidic backbone after digestion with EcoRI. The resulting expression cassette, flanked by AAV serotype 2 inverted terminal repeats (ITRs), contains the CAG promoter combining the cytomegalovirus early enhancer and the chicken β-actin promoter, a chicken β-actin intron, and a rabbit β-globin polyadenylation signal. rAAV2/1 and rAAV2/8 vectors were generated using a standard helper-virus free three-plasmid transient transfection method and pseudotyped with either AAV1 or AAV8 capsid proteins. Vectors were purified by two cesium chloride gradient centrifugations and dialyzed against PBS as described (Salvetti et al., 1998) . Genome titers of vector preparations were assayed by Dot-blot hybridization using a probe to detect the CAG or CMV promoter. Quantification of soluble Ova (sOVA) concentration in serum was performed by Ova-specific ELISA. Microtiter plates were coated with polyclonal rabbit anti-Ova antibodies (1:3000 dilution, Ray-Biotech) and bound sOVA was detected using biotinylated rabbit polyclonal anti-Ova antibodies (1:5000 dilution, Abcam) and streptavidin-peroxidase (1:15000, Roche). Detection of serum anti-Ova IgG antibodies was performed by ELISA using Ova-coated microtiter plates. Anti-Ova IgG antibodies were detected using biotinylated polyclonal goat anti-mouse IgG and revealed using the mouse ExtrAvidin kit (Sigma-Aldrich). IgG titers were defined as the dilution yielding the half maximum optical density obtained with control serums and was calculated using sigmoid curve fitting using GraphPad prism software. To analyze the quantities of Ova DNA and Ova mRNA present in transduced muscles of treated mice at indicated time points, a real-time PCR assay was developed. Muscles collected from each mouse were kept at −80˚C before DNA and RNA extraction performed using the phenol/chloroform method and the RNeasy Fibrous Tissue Mini Kit (Qiagen) following the manufacturer's instructions. For quantification of Ova DNA, the primers used were Ova-F (5 -AAG CAG GCA GAG AGG TGG TA-3 ), Ova-R (5 -GAA TGG ATG GTC AGC CCT AA-3 ), CD8a-F (5 -GGT GCA TTC TCA CTC TGA GTT CC-3 ), and CD8a-R (5 -GCA GAC AGA GCT GAT TTC CTA TGT G-3 ). For all reaction mixtures, 10 μl of FastStart Universal SYBR green master mix (Roche) was used in a final volume of 20 μl. Ova primers were used at 500 nM and CD8a primers at 400 nM. Approximately 10 ng of DNA was added in a 5-μl volume and always set up in duplicate. Each qPCR was performed under the following conditions: 10 min hot-start denaturation at 95˚C and 40 amplification cycles (10 s at 95˚C, Frontiers in Microbiology | Microbial Immunology 30 s at 60˚C). The melting temperatures of the final double-strand DNA products were determined by gradual heating from 60 to 95˚C over 20 min. All qPCRs were performed with a StepOne-Plus real-time PCR system (Applied Biosystems) and corresponding software. Absolute amounts of Ova and CD8a amplicons, in arbitrary units, were determined using serial dilutions of pAAV-CMV-OVA plasmid or pTOPO-CD8a plasmid as a standard. The data were expressed as Ova/CD8a ratios, fixed at 1 for PBS-injected control mice. For quantification of Ova mRNA, 100 ng of total RNA were reverse transcribed using iScript DNA Synthesis Kit (Biorad). Then, 2 μl of cDNA were subjected to real-time PCR amplification using Ova primers, β-actin-F (5 -AAG ATC TGG CAC CAC ACC TTC T-3 ) and β-actin-R (5 -TTT TCA CGG TTG GCC TTA GG-3 ) primers. For all reaction mixtures, 10 μl of FastStart Universal SYBR green master mix (Roche) was used in a final volume of 20 μl, Ova primers were used at 500 nM and β-actin primers at 400 nM. The same qPCR program as above-described conditions were used. The absolute amount of Ova mRNA for each sample was then normalized against the β-actin mRNA amount (arbitrary units) and determined using serial dilutions of pAAV-CMV-OVA plasmid and β-actin purified PCR product. Fluorescently labeled anti-CD4 (RM4-5), -CD8 (53-6.7), -CD44 (IM7), -CD62L (MEL-14), -PD-1 (J43), -PD-L1 (B7-H1), -PD-L2 (B7-DC) monoclonal antibodies (mAb), and unlabeled anti-CD16/CD32 antibodies were all purchased from eBioscience. PE-conjugated H-2K b /Ova 257-264 pentamers were used to detect the CD8 + cells that specifically recognize the immunodominant Ova-derived SIINFEKL peptide using the manufacturers' protocol (Proimmune). Flow cytometry measurements of single cell suspensions derived from lymph nodes, spleen, or blood samples were performed using standard procedures and acquired on a FACS-Canto (BD Biosciences) instrument. Flow cytometry analyses were performed using the FlowJo software (Tree Star). Enzyme-linked immunospot (ELISpot) assays were performed following the instructions of the manufacturer (Diaclone). Briefly, PVDF membrane plates were coated with capture antibody against mouse IFNγ and blocked with a 2% skimmed milk solution prepared in PBS. 2 × 10 5 Splenocytes per well were cultured in vitro for 36 h in the presence of 10 μg/ml SIINFEKL peptide. Plates were revealed after incubation with anti-mouse IFNγ detection antibody coupled to biotin followed with a streptavidinalkaline phosphatase conjugate and exposure to a ready-to-use solution of nitro-blue tetrazolium (NBT) and 5-bromo-4-chloro-3 -indolyphosphate (BCIP) for chromogenic development. Plates were analyzed with an ELISpot plate reader and a dedicated ImmunoSpots software (C.T.L.). Results were expressed as mean ± SEM. Significance was assessed by non-parametric one-way ANOVA (Kruskal-Wallis tests) using the GraphPad Prism software. Results were considered statistically significant when the p value was inferior to 0.05 (*), to 0.01(**) or to 0.001 (***). We used rAAV2/1-Ova and rAAV2/8-Ova administration to model a gene therapy setting where the transgene encodes for a highly immunogenic secreted protein. We first evaluated the capacity of these vectors to transduce muscle cells upon direct i.m. injection. Injection of 10 11 viral genome (vg) rAAV2/1-Ova or rAAV2/8-Ova similarly resulted in efficient transduction, as attested by the detection of Ova DNA and mRNA in the injected muscles quantified by qPCR and qRT-PCR respectively ( Figure 1A) . No significant difference was found when comparing the transduction efficiency of rAAV2/1-Ova and rAAV2/8-Ova in these conditions. Transgene expression in muscle was found to be high early after transduction but rapidly declined over time, suggesting the occurrence of an immune response against transduced cells ( Figure 1A) . To investigate this point, we quantified the cellular and humoral immune responses directed against the xenogenic Ova transgene product by flow cytometry and ELISA. Staining with MHC-I pentamers presenting the immunodominant Ova-derived SIINFEKL peptide, designated hereafter as H-2K b /Ova pentamers, allowed to monitor the expansion of CD8 + lymphocytes in spleen and draining lymph nodes of rAAV-Ova challenged animals. This analysis revealed that both rAAV2/1-Ova and rAAV2/8-Ova injected in the gastrocnemius muscles induced a robust expansion of anti-Ova CD8 + lymphocytes (Figures 1B,C) . Interestingly, the kinetics and intensity of T cell priming was found to be different when comparing the cells harvested from the draining lymph nodes and from the spleen. In the lymph nodes draining the injected muscles, anti-Ova CD8 + T cells expanded early after AAV challenge to reach around 2% of the CD8 + T cell compartment at day 7 post-injection ( Figure 1C , left panel). In contrast, when analyzing splenocytes, anti-Ova CD8 + T cells were found to reach up to 10% of the total CD8 + subset but only at day 14 after i.m. AAV injection ( Figure 1C , right panel). These results are consistent with a model in which the local cellular immune response measured in the draining lymph nodes would be induced by an rAAV leakage from the injected muscle and the consequent transduction of non-muscle cells such as dendritic cells, as shown for AAV5 (Xin et al., 2006) or AAV1 (Lu and Song, 2009 ) for instance. The systemic response detected in the spleen would rather reflect systemic immunization directed against the circulating sOva protein after it has been produced by transduced muscle cells. This interpretation is supported by the kinetics of sOva apparition and antibodies production ( Figure 1D) . Indeed, in the serum of AAVtreated mice, sOva reached a maximum at day 7 post-injection and rapidly declined thereafter ( Figure 1D , green curves). Consistently, anti-Ova IgG appeared between day 7 and 12 after AAV i.m. injection, correlating with the disappearance of sOva in the serum after day 7 ( Figure 1D , blue curves). The latter probably reflects the in vivo clearance of sOva following the formation of antigenantibody immune complexes, although we cannot completely rule out that anti-Ova IgG antibodies may mask Ova epitopes in the immunoassay. Together, these data reveal that i.m. injection of www.frontiersin.org rAAV encoding sOva is strongly immunogenic and provides a mouse model that is suited to stringently evaluate the efficacy of tolerance induction protocols. We investigated the possibility to block immune responses directed against the transgene product by two distinct immunomodulatory strategies, one blocking lymphocyte priming (i.e., CTLA-4/Ig) and the other inhibiting the function of lymphocytes targeting transduced muscle fibers (i.e., activation of the PD-1 pathway by muscle gene transfer of PD-1 ligands). Indeed, we have previously shown in transgenic mice that express Ova as a neo-autoantigen in muscle that tolerant transgenic anti-Ova CD8 + T cells up-regulated PD-1, suggesting that PD-1 plays a role in the induction of tolerance to skeletal muscle-expressed Ag (Calbo et al., 2008) . As expected, the anti-Ova CD8 + T cell response that followed an i.m. injection of rAAV2/1-Ova (Figures 1B,C) was accompanied by an up-regulation of PD-1 on CD8 + T cells that exhibit an activated CD44 hi phenotype (Figure 2A) . As activated T cells express PD-1 and PD-1 ligands inhibit their activation and function, we generated rAAV2/1-PD-L1 and rAAV2/1-PD-L2 vectors and first verified that they were able to efficiently transduce HEK cells in vitro ( Figure 2B) . The expected consequence of injecting these vectors i.m. is that they should not prevent T cell priming but rather inhibit the cytotoxic activity of T cells directed against muscle-expressed transgenic proteins. To investigate the in vivo immunomodulatory potential of CTLA-4/Ig and rAAV2/1-PD-L1, we either administered 200 μg of CTLA-4/Ig i.p. or co-injected rAAV2/1-PD-L1 i.m. at the same time as rAAV2/1-Ova, and monitored the cellular and humoral immune response 14 day after. Results showed that even a single dose of CTLA-4/Ig was efficient to completely prevent the priming of anti-Ova CD8 + T cells (Figure 2C , left panel) and the apparition of anti-Ova IgG in the serum of treated animals ( Figure 2C , middle panel). Accordingly, this immunosuppressive treatment allowed a sustained sOva production that could be assayed in the serum at day 14 whereas it was undetectable in controls ( Figure 2C right panel). We concluded from these data that transient immunosuppression by a single dose of CTLA-4/Ig is very efficient to prevent the priming of the cellular and humoral response at early time points after AAV transduction. In contrast, despite a slight but not statistically significant tendency to reduce immune responses, rAAV2/1-PD-L1 co-injected with rAAV2/1-Ova failed to provide the same effect on immune priming (Figure 2C , left and middle panels). Also, treatment with rAAV2/1-PD-L1 did not allow the persistence of sOva in the serum in agreement with the significant amount of anti-Ova IgG detected in the serum of these animals ( Figure 2C ). This expected result confirmed that local transduction of muscle cells with rAAV2/1-PD-L1 can not interfere with the systemic immune priming against sOva since activation of the PD-1/PD-L1 pathway was anticipated to regulate T cells functions rather than their priming. We further analyzed the presence of Ova DNA and mRNA in muscles at day 40 in the different groups of treated mice by qPCR and qRT-PCR, respectively. This further confirmed that suppression of immune priming by CTLA-4/Ig promotes the persistence of transduced cells in vivo as measured by the detection of significantly higher levels of Ova DNA and mRNA in the muscles of treated animals ( Figure 2D ). By contrast, injection Frontiers in Microbiology | Microbial Immunology Figure 1 with 1 × 10 11 vg rAAV2.1-Ova in the gastrocnemius muscles at day 0. Splenic CD8 + T cells were analyzed at day 14 by flow cytometry for expression of CD44 and PD-1. (B) rAAV2/1-PD-L1 and rAAV2/1-PD-L2 vectors were designed, produced, and tested for their capacity to transduce HEK-293 cells in vitro. For this, cells were analyzed by flow cytometry 3-5 days after their transduction with 1/10th dilution of the concentrated virus stocks. Controls correspond to unmanipulated HEK-293 parental cells stained with the same antibodies (green histograms). (C) The immunosuppressive potential of rAAV2/1-PD-L1 and CTLA-4/Ig were evaluated in vivo. Mice were injected with 10 11 vg rAAV2/1-Ova in the gastrocnemius muscles and received or not a co-injection of 10 11 vg rAAV2.1-PD-L1, or 200 μg of CTLA-4/Ig injected contemporaneously by the i.p. route. Blood samples were then collected 14 days later to analyze the percentage of anti-Ova CD8 + T cells, the level of anti-Ova IgG and the presence of sOva in the serum. (D) Gastrocnemius muscles were then collected at day 40, and Ova DNA and mRNA were quantified by qPCR and qRT-PCR. of rAAV2/1-PD-L1 did not significantly improve transgene persistence in these settings, in line with the prominent systemic immune response detected at day 14 ( Figure 2D ). Whereas a single dose of CTLA-4/Ig was efficient to suppress the systemic immune response at day 14 ( Figures 3A,B left panels), this was not the case when further monitoring the immune response at day 40. Indeed, anti-Ova CD8 + T cell expansion as well as anti-Ova IgG immune responses were readily detectable at day 40 in mice that had received rAAV2/1-Ova and CTLA-4/Ig at day 0 ( Figures 3A,B) . Therefore, early suppression of immune priming by a single dose of CTLA-4/Ig is not sufficient to permanently wipe out the immune responses against transgene product in these experimental conditions. Instead, continuous production of sOva by transduced muscle cells sensitizes the immune response at later time points when CTLA-4/Ig has been cleared from the circulation of treated mice. We next tested whether rAAV2/1-PDL-L1 or rAAV2/1-PDL-L2 could synergize with CTLA-4/Ig to improve transgene tolerance at later time points when the initial CTLA-4/Ig monotherapy was evidently not efficient alone to completely block immune sensitization against sOva. This combined strategy may indeed target two non-redundant mechanisms of immunomodulation acting www.frontiersin.org either on the APC side for CTLA-4/Ig or on the target tissue side for rAAV2/1-PD-L1 and rAAV2/1-PDL-L2. To test this possibility, groups of mice were transduced i.m. with rAAV2/1-Ova and also received at the same time either rAAV2/1-PD-L1 or rAAV2/1-PD-L2 i.m. and CTLA-4/Ig i.p. combination therapies. We then evaluated the cellular and humoral immune responses at different time points and finally evaluated the persistence of transduced muscle cells by quantification of Ova DNA and mRNA at day 80 (Figure 3) . Remarkably, whereas CTLA-4/Ig alone was inefficient, rAAV2/1-PD-L1 co-administered with CTLA-4/Ig at day 0 significantly improved transgene persistence, as measured by quantification of Ova DNA by qPCR (Figure 3C, left panel) . Accordingly, mRNA analysis by qRT-PCR at day 80 revealed sustained transcription of the transgene in muscle (Figure 3C, right panel) . Of note, rAAV2/1-PD-L2 appeared equally as effective as rAAV2/1-PD-L1 to provide this effect when combined with CTLA-4/Ig. Importantly, this protective effect occurred in a setting where neither CTLA-4/Ig alone nor the tested combination therapies could block immune response at days 40 and 80 (Figures 3A,B) . To further attest the presence of functional anti-Ova cytotoxic T cells at day 80 (a time point when pentamer staining is not sensitive enough to detect anti-Ova T cells, not shown), we analyzed by ELISpot the capacity of CD8 + T cells to secrete IFNγ when stimulated with the Ova-derived immunodominant SIINFEKL peptide. This assay revealed the presence of low numbers of functional anti-Ova CD8 + T cells in the three groups of animals that had received CTLA-4/Ig with or without rAAV-PD-L1 or rAAV-PD-L2, whereas the assay was negative for mice that did not receive rAAV-Ova and strongly positive for those that received rAAV-Ova in the absence of any immunoregulatory adjuvant therapy ( Figure 4A) . To definitively demonstrate that these IFNγ-secreting T cells were functional in vivo, we performed adoptive transfer experiments to non-lethally irradiated syngenic recipients that were inoculated with Ova-bearing tumor cells. Whereas tumors rapidly developed in recipients that had received T cells from control mice that only received PBS, recipient mice receiving lymphocytes from rAAV2/1-Ova injected donors rapidly developed an anti-tumor immune response that resulted in complete tumor rejection (Figure 4B ). In agreement with the lower numbers of anti-Ova CD8 + T cells detected by ELISpot, recipient mice that had received immune cells from donors treated with CTLA-4/Ig or the combined CTLA-4/Ig plus rAAV-PD-L1 or rAAV-PD-L2 therapy reacted with a slight delay but were also was also capable to efficiently control tumor growth ( Figure 4B) . Together, these results indicate that the cytotoxic activity of transgene product specific T cells that have escaped costimulation blockade can be functionally blocked by AAV-mediated gene transfer of PD-1 ligands in muscle tissues while they remain capable to reject tumor cells. Gene therapy for muscular dystrophies and other monogenic diseases aims at achieving long-term expression of a functional form of an otherwise deficient gene in target tissues. In this context, the product of the therapeutic gene is structurally different from its abnormal or absent counterpart, and consequently viewed as nonself by the immune system. Additional signals from the viral vector may further strengthen adverse innate and adaptive immune reactions against the transgene product, while the vector itself can be targeted by antiviral pre-existing or acquired immunity. Although the presence of natural epitopes in the non-mutated regions of the protein or the occasional occurrence of mutation reversion in a minute population of patients' cells (Klein et al., 1992) may yield some level of immunological tolerance, it remains that acquired immunity to the transgene product still represents one of the major obstacles to the success of gene therapy. Here, we provide the proof of principle that a combination immunoregulatory therapy targeting two non-redundant checkpoints of the immune response, i.e., priming and effector functions, can promote persistence of transduced target cells and transgene transcription thereof even when some cytotoxic T cells have escaped initial control. A first strategy for the blockade of adaptive immunity is obviously to target costimulation pathways and therefore provide an early control on lymphocyte priming. Whereas several costimulatory pathways have now been identified, CD28-mediated T cell costimulation by APC expressing CD80/CD86 molecules from the B7 family is clearly prominent. Indeed, CD28 signaling strongly enhances T cell proliferation and survival, cytokine production and prevents induction of anergy after TCR-mediated activation (Boise et al., 1995) . Consequently, anti-CD80 (B7.1) or anti-CD86 (B7.2) antibodies inhibit T cell priming and genetically engineered CD28-deficient T cells are strongly impaired in their capacity to expand (Green et al., 1994; Salomon and Bluestone, 2001) . T cells activation also promotes up-regulation of the negative regulator CTLA-4, another ligand of CD80 and CD86 molecules, which transmits inhibitory signals that regulate activated T cell and provides a key homeostatic mechanism of the immune response (Fife and Bluestone, 2008) . One of the early striking evidence of the key role of this molecule is that CTLA-4-deficient animal rapidly die FIGURE 4 | Detection of anti-Ova T cells in animals treated with combination therapies. Seven mice per group were injected with 1 × 10 11 vg rAAV2/1-Ova in the gastrocnemius muscles at day 0 and received at the same time combination therapies as in Figure 3 . Splenocytes were then harvested 80 days after and analyzed for the presence of lymphocytes capable to respond to Ova antigen. (A) Splenocytes were analyzed by ELISpot as described in the Section "Materials and Methods" for their capacity to secrete IFNγ after in vitro restimulation with the Ova-derived immunodominant SIINFEKL peptide. (B) 5 × 10 7 Splenocytes from the same mice were also adoptively transferred to 5 Gy-irradiated syngenic mice inoculated 1 day after with Ova-bearing EG7 tumor cells. The capacity of transferred lymphocytes to reject the tumors was monitored three times per week by measuring the tumor size with a digital caliper. www.frontiersin.org from massive lymphoproliferative syndrome and dysregulation of immunity (Waterhouse et al., 1995) . One convenient way to pharmacologically block the CD28 costimulation pathway is to use the soluble CD80/CD86 ligand CTLA-4/Ig. This fusion protein prevents CD28 interaction with CD80 and CD86 molecules on naïve and activated T cells and therefore provides a strong and early inhibition on T cell priming (June et al., 1990; Lenschow et al., 1992; Larsen et al., 2005) . CTLA-4/Ig may also induce a negative signalization on the APC side through the induction of indoleamine2,3-dioxygenase (IDO) which catalyses the production of inhibitory kynurenine from tryptophan (Grohmann et al., 2002) . CTLA-4/Ig may also possibly provide some level of APC depletion in vivo, although this aspect has been poorly documented and should critically depend on the form of CTLA-4/Ig used. Importantly, CTLA-4/Ig is now clinically available for the treatment of diseases such as rheumatoid arthritis (Dumont, 2004) and it is thus reasonable to propose its evaluation in the field of gene therapy. Indeed, there has been experimental evidence that CTLA-4/Ig in combination with a monoclonal antibody to CD40L (MR1) in the context of gene therapy can block immune responses and allow vector readministration in the mouse (Halbert et al., 1998; Lorain et al., 2008) . It is difficult to weight the relative roles of CTLA-4/Ig and MR1 in these experiments but a plausible hypothesis is that CTLA-4/Ig provides the strongest immunoregulatory contribution since MR1 alone has been reported to be less efficient in similar experimental conditions (Manning et al., 1998) . The results reported herein are consistent with these earlier reports and provide definitive evidence that CTLA-4/Ig alone efficiently, but only transiently, blocks anti-transgene T cell priming. One concern about using CTLA-4/Ig in gene therapy is that discontinuation of treatment exposes patients to immunization against the transgene product. The present results illustrate this point since initial CTLA-4/Ig efficiently blocks the anti-Ova cellular and humoral responses while sustained sOva production finally elicits this undesired response when CTLA-4/Ig is probably cleared from the circulation. Our experimental conditions have been purposely designed to model this very situation where the immune response escapes the initial immunoregulatory regimen in order to evaluate if a second strategy could provide additional protection of transduced cells, i.e., PD-L1/2 gene therapy. The PD-1/PD-L1 pathway has recently been identified as a negative regulator of immunity . The PD-1 immunoreceptor is also a member of the CD28/CTLA-4 family which, upon interaction with either one of its two ligands PD-L1 or PD-L2, down-modulates TCR signaling, reduces cytokine production and affects T cell survival by recruiting the SHP-2 tyrosine phosphatase, thereby dampening the PI3K and Akt pathway (Francisco et al., 2010) . In BALB/c mice, PD-1 deficiency leads to the spontaneous development of a lethal autoimmune cardiomyopathy . This disease is mediated by autoantibodies directed against the cardiac muscle autoantigen troponin I (Okazaki et al., 2003) . Hence, PD-1 controls the physiological tolerance to muscle autoantigens. Also, PD-L1 was recently shown to regulate CD8 + T cell-mediated muscle injury in a mouse model of myocarditis (Grabie et al., 2007) . Several PD-1/PD-L1 dependant mechanisms may act synergistically to induce an effective effector T cell blockade and contribute to tolerance induction. As mentioned earlier, PD-1 engagement on the surface of T cell is known to significantly inhibit TCR signaling thereby preventing T cell activation and cytokine production. In addition to this mechanism, PD-L1 was recently shown to be a potent inducer of adaptive Tregs. Indeed, PD-L1 expressed on CD8 + dendritic cell subset plays an important role on the conversion of naive lymphocytes toward FoxP3 + regulatory T cells (Wang et al., 2008) . Thus, engagement of PD-1 on T cells could induce cell-intrinsic tolerance (by blocking TCR signaling) as well as a dominant form of immunological tolerance by promoting the emergence of FoxP3 + adaptive Tregs. Of note, manipulation of the PD-1/PD-L1 axis is also used by some tumors cells to tolerize surrounding lymphocytes. Indeed, expression of PD-L1 on the surface of malignant cells has been suggested to represent a subversive strategy used by tumors to escape to immuno-surveillance . Thus, we believe that up-regulation of the expression of PD-L1 on the surface of muscle cells by the means of gene transfer should be able to induce tolerization of the immune system by preventing cytotoxicity of PD-1-expressing activated lymphocytes. However, PD-L1 gene transfer alone was not sufficient in our model to prevent the immune response directed against transduced muscle fibers (Figure 2) , suggesting that muscle expression of PD-L1, i.e., at distance from lymphoid compartments, is probably not capable of inducing sufficient adaptive FoxP3 + Tregs. Another explanation would be that PD-L1 expressed in muscle is not effective to block fully activated T cells that have received normal costimulatory signals (i.e., in the absence of CTLA-4/Ig co-administration). This interpretation would be in line with the finding that PD-1 inhibitory pathway could be overcome by CD28 costimulatory ligation in the presence of IL-2 (Freeman et al., 2000; Latchman et al., 2001; Carter et al., 2002; Ishida et al., 2002) . Hence, the use of CTLA-4/Ig to block early CD28 costimulatory signaling may create favorable conditions where escaping T cells could become sensitive to PD-L1/2. This strengthens the notion that CTLA-4 and PD-1 signaling represent two distinct pathways acting synergistically to maintain tolerance (Fife and Bluestone, 2008) . As shown herein, neither CTLA-4/Ig alone nor combination therapies could block immune response at days 40 and 80 (Figures 3 and 4) . Our results are thus compatible with the notion that the use of rAAV2/1-PD-L1 or rAAV2/1-PD-L2 in combination with CTLA-4/Ig, does not completely inhibit the immune priming against sOva that most probably occurs at distance from the site of transduction (e.g., spleen and/or lymph nodes) but instead provides an additional protection by disarming lymphocytes within the target tissue. It has been argued that the priming of T cells directed to the transgene product may be somehow defective after rAAVmediated gene transfer (Lin et al., 2007a,b) . However, in our experimental settings, transduced muscle fibers progressively disappear in the absence of immunoregulatory treatment ( Figure 2D) . Further, we show that even after CTLA-4/Ig alone or in combination therapies, an anti-Ova immune response is evidenced in ELISpot assays and in an in vivo model of tumor rejection upon adoptive transfer of T cells in recipient animals (Figure 4) . Hence, forced up-regulation of PD-1 ligands in muscle cells yielded a level of protection of transduced cells from the cytotoxic assault Frontiers in Microbiology | Microbial Immunology of circulating anti-Ova lymphocytes and provided in combination therapy a means to prolong transgene persistence and transcription. Therefore, CTLA-4/Ig plus PD-L1/2 combination therapy represents a candidate approach to circumvent the bottleneck of immune responses directed toward the transgene product and may deserve further investigation for non-secreted transgenic protein models and in larger animals. Sahil Adriouch designed research, performed experiments, analyzed data, and co-wrote the manuscript; Anna Salvetti and Olivier Boyer designed research and co-wrote the manuscript; and Emilie Franck, Laurent Drouot, Carole Bonneau, and Nelly Jolinon performed experiments.

Comparison of Humoral Immune Responses to Epstein-Barr Virus and Kaposi’s Sarcoma–Associated Herpesvirus Using a Viral Proteome Microarray

Background. Epstein-Barr virus (EBV) is a ubiquitous herpesvirus, and Kaposi’s sarcoma–associated herpesvirus (KSHV) has a restricted seroprevalence. Both viruses are associated with malignancies that have an increased frequency in individuals who are coinfected with human immunodeficiency virus type 1 (HIV-1). Methods. To obtain an overview of humoral immune responses to these viruses, we generated a protein array that displayed 174 EBV and KSHV polypeptides purified from yeast. Antibody responses to EBV and KSHV were examined in plasma from healthy volunteers and patients with B cell lymphoma or with AIDS-related Kaposi’s sarcoma or lymphoma. Results. In addition to the commonly studied antigens, IgG responses were frequently detected to the tegument proteins KSHV ORF38 and EBV BBRF and BGLF2 and BNRF1 and to the EBV early lytic proteins BRRF1 and BORF2. The EBV vIL-10 protein was particularly well recognized by plasma IgA. The most intense IgG responses to EBV antigens occurred in HIV-1–positive patients. No clear correlation was observed between viral DNA load in plasma and antibody profile. Conclusions. The protein array provided a sensitive platform for global screening; identified new, frequently recognized viral antigens; and revealed a broader humoral response to EBV compared with KSHV in the same patients.

Background. Epstein-Barr virus (EBV) is a ubiquitous herpesvirus, and Kaposi's sarcoma-associated herpesvirus (KSHV) has a restricted seroprevalence. Both viruses are associated with malignancies that have an increased frequency in individuals who are coinfected with human immunodeficiency virus type 1 (HIV-1). Methods. To obtain an overview of humoral immune responses to these viruses, we generated a protein array that displayed 174 EBV and KSHV polypeptides purified from yeast. Antibody responses to EBV and KSHV were examined in plasma from healthy volunteers and patients with B cell lymphoma or with AIDS-related Kaposi's sarcoma or lymphoma. Results. In addition to the commonly studied antigens, IgG responses were frequently detected to the tegument proteins KSHV ORF38 and EBV BBRF and BGLF2 and BNRF1 and to the EBV early lytic proteins BRRF1 and BORF2. The EBV vIL-10 protein was particularly well recognized by plasma IgA. The most intense IgG responses to EBV antigens occurred in HIV-1-positive patients. No clear correlation was observed between viral DNA load in plasma and antibody profile. Conclusions. The protein array provided a sensitive platform for global screening; identified new, frequently recognized viral antigens; and revealed a broader humoral response to EBV compared with KSHV in the same patients. Epstein-Barr virus (EBV) and Kaposi's sarcomaassociated herpesvirus (KSHV) are human herpesviruses that differ dramatically in their geographic prevalence. EBV seropositivity in healthy populations is .90% worldwide. KSHV seropositivity is low (0-5%) in the United States, Canada, and Europe; intermediate (7-24%) in Italy and the Mediterranean; and high (23-70%) in sub-Saharan Africa [1] [2] [3] [4] . Seroprevalence for KSHV increases in individuals coinfected with human immunodeficiency virus 1 (HIV-1) and, in the United States and Europe, in men who have sex with men [5] [6] [7] . Primary infection with EBV can cause infectious mononucleosis, and both EBV and KSHV are associated with human cancers. EBV is associated with nasopharyngeal carcinoma, African Burkitt's lymphoma, a subset of Hodgkin's lymphoma, AIDS-related lymphoma, gastric carcinoma, peripheral T cell lymphoma, and posttransplant and other lymphoproliferative diseases in the immunodeficient [8] . KSHV is associated with Kaposi's sarcoma, primary effusion lymphoma, and multicentric Castleman's disease. Immunosuppression, such as that imposed by concurrent HIV-1 infection or transplantation regimens, increases the incidence of some of these virus-associated cancers [8, 9] . The DNA genomes of EBV and KSHV encode an estimated 80 and 86 proteins, respectively. The degree to which these proteins are expressed differs depending on the stage of the viral life cycle. In vivo, the reservoir of EBV latent infection is resting memory B cells that have either no EBV protein expression or transient expression of the EBNA1 protein. Expression of the other latency proteins, EBNA2, EBNA3A, EBNA3B, EBNA3C, EBNA-LP, LMP-1, and LMP2A and LMP2B, has been detected in peripheral blood B cells shortly after primary infection and in B cells in tonsillar tissues [10, 11] . Lytic replication and expression of the EBV lytic proteins takes place in plasma cells and in epithelial cells [12] [13] [14] . Less is known about in vivo persistence of KSHV. B cells form a reservoir for latent KSHV infection, and differentiation of these cells also triggers lytic KSHV induction [15, 16] . KSHV endothelial and B cell tumors express the LANA, v-cyclin, and v-FLIP latency proteins, and vIRF3 is also detected as a latency protein in primary-effusion lymphoma cells. In addition, a small percentage of tumor cells express lytic cycle proteins such as the v-IL6 and the viral G-protein coupled receptor (v-GPCR), suggesting that KSHV lytic infection may contribute to the tumorigenic phenotype [17] . Transcriptional profiling arrays have been developed to examine genome-wide expression of EBV and KSHV [18, 19] . However, humoral immune responses to EBV and KSHV proteins have predominantly been measured using virus-infected cell lines or enzyme-linked immunosorbent assays that incorporate just a few selected viral proteins [2] [3] [4] [20] [21] [22] . Although these assays have proven valuable for diagnostic and epidemiological studies, they do not provide a complete picture of the full repertoire of antibody responses generated by EBV and KSHV infection. High-throughput technologies allow cloning and expression of entire pathogen proteomes [23, 24] . We used this technology to generate proteomic microarrays for EBV and KSHV. These arrays allowed us to compare the global humoral immune responses to EBV vs KSHV in the same patient samples and within the same assay. EBV and KSHV open reading frames (ORFs) were cloned and expressed as previously described [25] . Briefly, the open reading frames were polymerase chain reaction (PCR) amplified using primers based on the GenBank sequences V01555, AJ507799, and U756981. The bacterial artificial chromosome plasmids Akata BXI (EBV) and BAC36 (KSHV) [26] were used as templates (gifts from L. Hutt-Fletcher and S-J Gao). PCR products were cloned into the vector pDONR201. ORFs representing different virus strains were amplified directly from clinical samples or from cell lines. Proteins containing extensive repeats that limited expression in yeast (eg, EBNA1 and LANA) were amplified as N-terminal and C-terminal fragments. Escherichia coli bacteria were transformed with the reaction products and DNA prepared for sequence verification. Correct ORF-containing plasmid DNAs were then moved into a yeast destination vector, pEGH-A, a derivative of the yeast glutathione S-transferase (GST) vector, pEGH. Yeast cultures were induced with 2% galactose, and GST fusion proteins were isolated and purified on glutathione beads. One hundred and seventy-four appropriately sized EBV and KSHV proteins (Table S1 ; available online) were successfully purified based on immunoblot analysis using anti-GST antibody and were printed using a 48-pin contact printer (Bio-Rad) in duplicate on modified glass (Full Moon Biosystems) microscope slides along with controls. Each slide had 784 spots: 348 EBV and KSHV proteins, 112 control proteins (Table S1 ; available online), and 324 blank spots. The controls were used for orientation and to detect nonspecific interaction with proteins such as the GST fusion partner. Protein arrays were stored at 280°C. Plasma from healthy blood donors and from patients with follicular B cell lymphoma (without HIV-1), AIDS-related lymphoma (ARL), and AIDS Kaposi's sarcoma was obtained with written informed consent in accordance with the Declaration of Helsinki after approval by the Johns Hopkins institutional review board. Plasma was separated from peripheral blood collected in standard EDTA or acid citrate dextrose tubes and stored at 280°C. Protein arrays were preincubated in 4 mL Superblock (Pierce) plus 2% bovine serum albumen (BSA) at 4°C overnight. Slides were assembled with 4-well modules and incubated with plasma (300 uL diluted in phosphate buffered saline [PBS] plus 2% BSA) for 1 hour at room temperature. Slides were washed with prewarmed (42°C) PBS Tween-20 (PBST) and incubated with secondary Cy3labeled antihuman IgG or IgA antibody (1:2000; Sigma and Jackson, respectively) diluted in PBS plus 2% BSA for 1 hour at room temperature. Slides were then washed 23 with prewarmed PBST, 13 with prewarmed distilled water, dried, scanned, and analyzed with GenePix Pro software (Molecular Devices). The software GenePix Pro 7 was used to obtain the median foreground and background intensity for each spot. The raw intensity of the spot was defined as the ratio of the foreground to background median intensity. There are 324 ''blank'' spots on each protein chip, and the raw intensity distribution of these spots has a mean value of approximately 1. Assuming that the raw intensity distributions of ''blank'' spots are the same across all microarrays, we can standardize the protein signals such that I2m r where Z is Z-score of each spot, I is raw intensity of the spot, and m and r are mean value and standard deviation, respectively, of ''blank'' spots on the microarray. Each protein has 2 duplicated spots. The identified hits were defined as those proteins with Z-scores for both spots at or above the 5 SD cutoff. When comparing the appearance frequency for individual antigens for HIV/KS patients vs lymphoma or healthy normals, the enrichment score was reflected by ES 5 f ks =f 1 . We adopted binominal probability to calculate the significance between comparisons such that where p is the probability that the protein was hit for at least k HIV/KS patients, n is the total number of HIV/KS patients, and f 1 is the frequency of the protein in the comparison group (either lymphoma or healthy normals). Proteins were selected as a significant biomarker for HIV 1 /KS patients only if they had at least 50% frequency in HIV 1 /KS patients, the enrichment score ES was larger than 1.6, and the P value was less than .005. EBV and KSHV DNA in plasma were detected as previously described [27] . Viral DNA levels were determined by real-time, quantitative PCR using K8 primers for KSHV and BamHI-W primers for EBV. Controls were constructed by the addition of viral DNA to serum from healthy donors. Standard curves (with duplicate serial 10-fold dilutions of plasmid DNA that included a target sequence from 10 5 to 10 copies) were run in parallel with each analysis. To systematically compare EBV and KSHV antibody responses, we utilized proteomic arrays displaying 174 virus proteins plus controls. The printing quality and the quantity of the immobilized proteins on the chip were monitored using anti-GST antibody followed by Cy3-labeled secondary antibody ( Figure 1A ). Preliminary analyses with arrays containing the 82 EBV and 92 KSHV ORFs revealed that plasma diluted up to 1:10 000 gave a signal on the arrays that was readily detected with Cy3-tagged antihuman IgG. A positive signal was set as one that was 5 SD or more above background. A section of the array illustrating positive and control signals is shown in Figure 1B . ORF38 is a tegument protein that has not been widely examined as an antigen. ORF73 encodes the latency LANA protein that is a gold standard for KSHV serology studies. The central region of LANA contains repeated sequences that limit effective expression of the protein in yeast. The N-terminus and C-terminus of LANA were therefore expressed separately for presentation on the array. The LANA C-terminus was recognized more frequently than the N-terminus (14 vs 4 positive plasma). This is consistent with a peptide mapping study that found more sera reacting to peptides mapping to the LANA C-terminus than to the N-terminus [28] . The median number of KSHV antigens recognized by the LANA-positive HIV 1 /KS plasma at 1:10 000 serum dilution was 6. Analysis of the EBV proteins recognized in the same assay by the same plasma samples revealed a different picture. EBV is a ubiquitous virus, and EBV proteins on the array were recognized using plasma from healthy donors, HIV 2 lymphoma patients, and HIV 1 /KS patients ( Figure 2B ). The median number of EBV proteins recognized at 1:100 serum dilution was 17, 18, and 27 for plasma from healthy donors, lymphoma patients, and HIV 1 /KS patients, respectively. Ninety-one percent of plasma specimens tested (32/35) reacted with the carboxy terminus of EBNA1 and 97% (34/35) reacted with the combination of EBNA3B and EBNA3C at 1:100 or 1:10 000 dilution. The single EBNA3-negative plasma specimen was from a lymphoma patient whose plasma did not recognize any EBV antigens. The difference between the normal and HIV 2 lymphoma population and the HIV 1 /KS population lay not in the individual antigens recognized, but rather in the heightened antigenic response to a range of EBV proteins. This is illustrated by the increased median number of antigens detected at 1:10 000 plasma dilution seen with HIV 1 /KS samples (15) vs the median number with plasma from healthy donors and lymphoma patients (7 and 5, respectively). Antigens with a statistically significant difference in appearance between healthy donors and HIV 1 /KS patients at a 1:10 000 dilution are shown in Figure 3 , which also includes the data for HIV-negative lymphoma patients for comparative purposes. The BFRF3 (p18) capsid antigen has been previously advocated as a serological marker for nasopharyngeal carcinoma diagnosis [29] . Two subtypes of EBV, type 1 and type 2 (also called A and B), were originally defined on the basis of differences in the EBNA2 gene [30] . In this set, 10/35 sera showed discordant responses to the 2 EBNA2 proteins with either a different sensitivity of detection (1:100 vs 1:10 000) or recognition of only 1 of the 2 EBNA2 proteins. It is likely that these individuals are infected with only 1 of the 2 EBV subtypes. Note that 9/10 discordant sera had preferential recognition of type 1 EBNA2, which is consistent with the greater prevalence of type 1 EBV in clinical samples from the United States. Positivity to both proteins would represent recognition of common epitopes or reflect concurrent infection with both virus subtypes. In contrast to the antibody responses to KSHV proteins, the profiles of antibody responses to EBV proteins in these plasma samples were similar in the KS and lymphoma-carrying HIVpositive populations ( Figure 4B ). There was no significant difference in either the intensity of the positive signals or the number of antigens being recognized at a 1:10 000 serum dilution. Thus, the presence of KSHV-associated disease did not appear to have any impact on humoral responses to EBV antigens in the setting of concurrent HIV-1 infection. Again, the broader antigenic response to EBV compared with KSHV is striking. The W1 exon of EBV BamHI-W encodes 22 amino acids that form part of the EBNA-LP protein [33, 34] . The protein encoded by the entire EBV BamHI-W ORF (BWRF1) and the EBNA-LP protein were both printed on the array. We noted that 23/24 (95%) of the specimens that recognized EBNA-LP also recognized the protein encoded by BWRF1, implying that the common 22 amino acids contribute significantly to antibody recognition of EBNA-LP. EBV and KSHV DNA are detected infrequently in serum or plasma from healthy individuals, but there is an increased prevalence in cancer patients and in HIV-1-infected individuals [35, 36] . A consistent correlation between EBV or KSHV DNA levels in serum or plasma and antibody titers to viral proteins has not been apparent [37] . To see if the more global antibody responses detected on the arrays would provide any additional insight, we examined KSHV and EBV DNA levels in the plasma from HIV-positive KS patients (see Figure 2A ). The plasma copy number for KSHV and EBV DNA was measured by real-time quantitative PCR using previously published primers [27] . A comparison of DNA copy number with antibody responses did not reveal any obvious correlation for either KSHV or EBV ( Figure 5) . However, the 2 samples with the highest KSHV DNA copy number (.180 000 copies) did not recognize either the K8.1 or ORF38 lytic proteins, whereas 61% (8/13) of the samples with ,50 000 copies recognized both of these antigens. Two samples are insufficient to allow any conclusions to be drawn, but the observation does raise the possibility that inadequate antibody production may have contributed to the robust KSHV replication detected in these individuals. IgA responses to EBV EBNA1 and viral capsid antigens have long been used as a diagnostic tool for nasopharyngeal carcinoma [38] . In addition to nasopharyngeal carcinoma, IgA responses to VCA and to EBNA1 are frequently elevated in lymphoma patients and in individuals who are HIV-1 positive [39] . IgA responses to EBV proteins were examined at a 1:1000 dilution for 15 of the HIV-positive KS patients, 10 healthy individuals, and 10 HIV-positive lymphoma patients. Seven EBV antigens were consistently recognized by IgA across these patient groups, namely, EBNA1, the helicase BBLF4, the viral IL10 BCRF1, the uracil DNA glycoslyase BKRF3, the early protein BRRF1, the tegument protein BRRF2 [40] , and the latency membrane protein LMP2A ( Figure 6 ). The most intense IgA response detected was to EBNA1, which appeared to be recognized even more readily by IgA than IgG. All 25 normal and HIV-positive KS plasma tested recognized EBNA1 at a 1:1000 dilution, whereas in 4 of these samples there was no recognition of EBNA-1 by IgG even at a 1:100 dilution. However, the IgA response to the functionally equivalent protein in KSHV, LANA, was very different. Of 14 HIV-positive KS plasma samples that had IgG responses to LANA at a 1:10 000 dilution, only 2 had IgA responses at a 1:1000 dilution (data not shown). In 1 previous study, IgA antibody to KSHV LANA was detected in saliva but not in serum [41] . A comparison of the IgG and IgA responses in the plasma from healthy donors also revealed differential responses to certain EBV proteins. All 10 normal sera had IgA responses at a 1:1000 dilution to the viral IL10 BCRF1, the tegument protein BRRF2, and the latency membrane protein LMP2A. In contrast, even at a 1:100 dilution, IgG responses to BCRF1, BRRF2, or LMP2A were not detected in these plasma samples. The IgA response to vIL10 is particularly interesting because vIL10 is an immune modulator that can have an impact on localized immune escape of EBV-infected cells. Testing for EBV and KSHV has relied on assays such as immunofluorescence staining and Western blotting of infected cells. Assays using individual viral proteins have also been described. However, in the absence of a global comparison of protein immunogenicity, there has been inadequate information on which to base the choice of antigens to incorporate into these assays. The EBV latency proteins EBNA1, EBNA2, and EBNA3C and the KSHV LANA latency and K8.1 lytic proteins have been widely used as antigens, and these antigens were also frequently detected on the arrays. Our protein array screens also identified the EBV tegument proteins BBRF2, BGLF2, and BNRF1 (FGARAT); the small capsid protein BFRF3; the early protein BRRF1; and the KSHV ORF38 tegument protein as lytic viral proteins that are particularly well recognized by IgG. The screens also identified BBLF4, BCRF1, BKRF3, and BBRF1 as EBV lytic proteins that were well recognized by IgA. It should be noted that the EBV and KSHV proteins for our array were expressed in yeast which do not have the same machinery as mammalian cells for complex glycosylation. It is therefore likely that our assays underestimate the immunological response to glycosylated viral antigens such as the EBV envelope glycoprotein gp350/220 and KSHV K8.1. Globally, the array analyses revealed a stark contrast between antibody responses to EBV and KSHV in the same patients. Multiple EBV proteins were consistently recognized by plasma IgG at the dilutions examined, whereas only the LANA (ORF73), ORF38, and K8.1 proteins of KSHV were detected with any consistency, even in patients with KS. This difference carried over to serum IgA responses where 2 latency EBV proteins and 5 lytic proteins were detected by the majority of the sera, while the only KSHV antigen detected by IgA was LANA, and that was seen only rarely. These observed differences may reflect the biology of in vivo persistence. Both viruses latently infect B cells, and both viruses are secreted into saliva [42, 43] . However, EBV can be routinely detected in circulating latently infected B cells of healthy seropositive individuals, whereas detection of KSHV in blood is rare [44] [45] [46] . The preferential ability to detect serum IgA responses against EBV vs KSHV on the protein arrays may also be related to in vivo infection of mucosal epithelial cells by EBV because the most frequently observed antigens were latency proteins or lytic replicative proteins known to be expressed in epithelial cells [47] [48] [49] [50] . KSHV infects endothelial cells as well as B cells, but the contribution of endothelial cells to in vivo viral persistence is not well understood. The protein array described here provided a highly sensitive platform. Plasma with EBNA1 titers between 1:80 and 1:320 in conventional anticomplement immunofluorescence assays had titers of 1:10,000 or greater in the array assay (data not shown). The assay showed specificity. IgG responses to KSHV proteins were detected in sera from HIV-positive individuals with a diagnosis of KS or lymphoma but were not detected in a panel of HIV-negative lymphoma patients or healthy normals. This result, along with comparisons between EBV and KSHV responses in the HIV-positive samples, indicates that cross-reactivity between homologous proteins encoded by EBV and KSHV was not a confounding factor. This array provides a new platform for investigating the contribution of humoral immune responses in patients with EBV-and KSHV-associated pathologies. Supplementary materials are available at The Journal of Infectious Diseases online (http://www.oxfordjournals.org/our_journals/jid/). Supplementary materials consist of data provided by the author that are published to benefit the reader. The posted materials are not copyedited. The contents of all supplementary data are the sole responsibility of the authors. Questions or messages regarding errors should be addressed to the author. Figure 6 . IgA responses to EBV proteins. Plasma specimens from HIV-1-positive KS patients, AIDS-related lymphoma patients, and healthy individuals were incubated on the protein arrays at a 1:1000 dilution (upper panel). Heat map showing responses detected with Cy3-labeled anti-human IgA that were above the 5 SD cutoff (lower panel). Number of antigens recognized by each sample. Financial support. This work was supported by Public Health Service grants from the National Cancer Institute of the National Institutes of Health (R37 CA42245 and R01 CA30356 to S. D. H, R21 CA138163 to S. D. H. and H. Z., RC2 CA148402 to G. S. H., and Cancer Center Core Grant P30CA006973 to William Nelson). Potential conflicts of interest. All authors: No reported conflicts.

Stem–loop structures can effectively substitute for an RNA pseudoknot in −1 ribosomal frameshifting

−1 Programmed ribosomal frameshifting (PRF) in synthesizing the gag-pro precursor polyprotein of Simian retrovirus type-1 (SRV-1) is stimulated by a classical H-type pseudoknot which forms an extended triple helix involving base–base and base–sugar interactions between loop and stem nucleotides. Recently, we showed that mutation of bases involved in triple helix formation affected frameshifting, again emphasizing the role of the triple helix in −1 PRF. Here, we investigated the efficiency of hairpins of similar base pair composition as the SRV-1 gag-pro pseudoknot. Although not capable of triple helix formation they proved worthy stimulators of frameshifting. Subsequent investigation of ∼30 different hairpin constructs revealed that next to thermodynamic stability, loop size and composition and stem irregularities can influence frameshifting. Interestingly, hairpins carrying the stable GAAA tetraloop were significantly less shifty than other hairpins, including those with a UUCG motif. The data are discussed in relation to natural shifty hairpins.

Ribosomal frameshifting is a translational recoding event in which a certain percentage of ribosomes are forced to shift to another reading frame in order to synthesize an alternative protein. This switch occurs at a specific position on the mRNA, called the slip site or slippery sequence, and can be either forwards (+1) or backwards (À1). The nature and efficiency of frameshifting depends on several factors, including tRNA availability and modifications, and mRNA primary and secondary structure (1, 2) . The signals that are responsible for À1 frameshifting comprise two elements: a slippery sequence where the actual reading shift takes place, and a downstream located structural element which greatly stimulates the efficiency of frameshifting. Although the mechanism is still elusive, the present view is that the downstream structure forms a physical barrier that blocks EF-2 function and causes ribosomes to stall in their translocation step. This 'roadblock' puts tension on the mRNA-tRNA interaction. The tension can be relieved by the realigning of A-site and P-site tRNAs in the 5 0 -direction, whereafter EF-2 can do its work and the ribosome resumes translation in the À1 reading frame (3) . In general, a pseudoknot is more efficient in stimulating frameshifting than a hairpin of the same sequence composition. This difference is likely related to a higher thermodynamic stability of the pseudoknot. Indeed, from thermodynamic analysis it appears that pseudoknots are more stable than their hairpin counterparts (4) (5) (6) . Recent studies employing mechanical 'pulling' of frameshifter pseudoknots have shown a correlation between the mechanical strength of a pseudoknot and its frameshifting capacity (7, 8) , and the influence of major groove and minor groove triplex structures (9) . The higher strength of a pseudoknot can be primarily attributed to the formation of base triples between the lower stem S1 and loop L2 ( Figure 1A ), making it more resistant against unwinding by an elongating ribosome (8, 10) . Base triples in several pseudoknots, such as Beet western yellows virus (BWYV) p1-p2 (11) , Pea enation mosaic virus type-1 (PEMV-1) p1-p2 (6) , Sugarcane yellow leaf virus (ScYLV) p1-p2 (12) and Simian retrovirus type-1 gagpro (SRV-1) (13, 14) have been shown to play an essential role in frameshifting. For pseudoknots with a longer stem S1 of 10-11 bp, like that of Infectious Bronchitis Virus (IBV), base triples do not appear to contribute to frameshifting (15) . Although a hairpin is considered to be a less efficient frameshift-inducing secondary structure than a pseudoknot, some viruses like Human immunodeficiency virus (HIV) (16) , Human T-lymphotropic virus type-2 (HTLV-2) (17) and Cocksfoot mottle virus (CfMV) (18) make use of a simple hairpin to stimulate substantial levels of frameshifting. In addition, frameshifting in the prokaryotic dnaX gene requires, next to an upstream enhancer, the presence of a hairpin as well (19) . A few studies have investigated a correlation between hairpin stability and frameshift efficiency of natural shifty hairpins (19, 20) . Nonetheless, certain studies have shown that a hairpin composed of the same base pairs as a frameshifter pseudoknot is not very efficient in inducing frameshifting in mammalian cells and lysates (21) (22) (23) but is in other systems (24) . Here, we have carried out a systematic analysis of the frameshift-inducing efficiency of hairpins derived from the SRV-1 gag-pro frameshifter pseudoknot. Investigation of about 30 different hairpin constructs revealed that next to thermodynamic stability, also loop size and composition, and stem irregularities can significantly influence frameshifting. Our data showed that there exists no base specific contacts between the hairpin and the ribosome during frameshifting and suggests that the hairpin primarily serves as a barrier to allow repositioning of tRNAs at the slippery site. Mutations in the SRV-1 gag-pro frameshifting signal were made in an abridged version of plasmid SF2 (25) which is derivative of pSFCASS5 (26), a frameshift reporter construct. In this version, the entire BglII-NcoI fragment of pSF2 was replaced by a synthetic dsDNA fragment (5 0 -G ATCTTAATACGACTCACTATAGGGCTCATTTAA ACTAGTTGAGGGGCCATATTTCGC-3 0 , a SpeI restriction site is underlined). This yielded plasmid pSF208 in which the original GGGAAAC slippery sequence has been replaced by the more slippery UUUAAAC sequence (26) . pSF208 was digested with SpeI and NcoI, and sets of complementary oligonucleotides corresponding to the various mutants were inserted. A list of oligonucleotides is available upon request. All constructs were verified by automated dideoxy sequencing using chain terminator dyes (LGTC, Leiden). DNA templates were linearized by BamHI digestion and purified by successive phenol/chloroform extraction and column filtration (Qiagen, Benelux). SP6 polymerase directed transcriptions were carried out in 50 ml reactions containing $2 mg linearized DNA, 10 mM NTPs, 40 mM Tris-HCl (pH 7.9), 10 mM NaCl, 10 mM DTT, 6 mM MgCl 2 , 2 mM spermidine, 6 U of RNase inhibitor (RNAsin, Promega, Benelux) and 15 U of SP6 polymerase (Promega, Benelux). After an incubation period of 2 h at 37 C, samples were taken and run on agarose gels to determine the quality and quantity of the transcripts. Appropriate dilutions of the reaction mix in water were directly used for in vitro translations. Alternatively, transcripts were purified by phenol/chloroform extraction and isopropanol precipitation and quantified by UV absorption as described previously (14) . Experiments were carried out in duplicate using seriallyin water-diluted mRNAs with final concentrations of 5 nM. Reactions contained 4 ml of an RNA solution, 4.5 ml of rabbit reticulocyte lysate (RRL, Promega), 0.25-1 ml of 35 S methionine (Amersham, in vitro translation grade), 0.5 ml of 1 mM amino acids lacking methionine and were incubated for 60 min at 28 C. Samples were boiled for 3 min in 2Â Laemmli buffer and loaded onto 12% SDS polyacrylamide gels. Gels were dried and exposed to phosphoimager screens. Band intensity of 0-frame and À1 frameshift products was measured using a Molecular Imager FX and Quantity One software (Biorad). Frameshift percentages were calculated as the amount of À1 frameshift product divided by the sum of 0 and À1 frame products, corrected for the number of methionines (10 in the 0-frame product and 28 in the fusion product), multiplied by 100. Candidates of interest were constructed in a dual luciferase vector, pDUAL-HIV(0), essentially as described previously (14, 27) . In short, pDUAL-HIV(0) was digested by KpnI and BamHI, followed by insertion of complementary oligonucleotides to clone the SRV-1 gag-pro pseudoknot, various hairpins as shown in Figures 2C and 5, and a negative control (NC) which formed no apparent secondary structure downstream of the slippery sequence. An in-frame control was constructed by inserting an A-residue upstream of the cytosine in the UUUAA AC slippery sequence of a 12 bp hairpin frameshift construct. HeLa cells were cultured in DMEM/high glucose/ stable glutamine (PAA Laboratories GmbH, Germany) and supplemented with 10% fetal calf serum and 100 U/ml penicillin and 100 mg/ml streptomycin. Cells were kept in a humidified atmosphere containing 5% CO 2 at 37 C. Assay protocols were described previously (14) . Briefly, cells were transfected with 300 ng of plasmid using 1 ml of lipofectamine-2000 (Invitrogen) in a 24-well plate. Cells were lysed 24 h after transfection and luciferase activities were quantified by Glomaxmultidetector (Promega, Benelux) according to manufacturer's protocol. Frameshifting efficiency was calculated by dividing the ratio of Renilla luciferase (RL) over Firefly luciferase (FL) activity of the mutant by the RL/ FL ratio of the in-frame control, multiplied by 100. In contrast to earlier reports involving the IBV frameshifting pseudoknot (21, 22) , we found that in the case of the SRV-1 gag-pro frameshift inducing pseudoknot a hairpin of similar composition as the pseudoknot did stimulate frameshifting in vitro ( Figure 1A and B). The 12 bp hairpin derivative of the SRV-1 pseudoknot (SRV-hp) showed 22% frameshifting efficiency, whereas the SRV-1 pseudoknot (SRV-pk) in this context yielded 31%. The pseudoknot in these experiments is a modified version of the wild-type SRV-1 pseudoknot previously used for NMR and functional analysis (14) . We note that the U UUAAAC slippery sequence was used to enhance the sensitivity of the in vitro frameshifting assay. This sequence is $1.5-fold more slippery than the wild-type GGGAAAC slippery sequence (28) . In the latter context, the hairpin was indeed less efficient (data not shown) while a non-slippery variant, GGGAAGC, was not effective at all (<0.2%, data not shown). Two other known efficient slip sites, AAAAAAC and UUUUUUA, caused 23 and 27%, respectively, of ribosomes to switch frame in the presence of the 12 bp hairpin (data not shown). These data showed that the 12 bp hairpin is a genuine stimulator of frameshifting. Since the hairpin construct also contained sequences resembling those of L2 of the pseudoknot construct, it was theoretically possible that these nucleotides could take part in the same base triples. To investigate this possibility, we replaced the downstream sequence in the hairpin construct (SRV-muthp). This did not affect the frameshift efficiency of the hairpin construct. In contrast, the same mutations in the pseudoknot context (SRVmutpk) reduced its activity about 6-fold ( Figure 1B) . Thus, it is unlikely that triple helix formation or other tertiary interactions contribute to hairpin-dependent frameshifting; the hairpin as such seems to be sufficient. Next, we investigated the role of stem length on frameshifting efficiency. Increasing stem size from 12 to 15 or 21 bp did not significantly alter frameshifting (Figure 2A ). On the other hand, decreasing stem size led to a steady decrease in frameshifting efficiency which seemed to vanish around a stem size of 5 bp or ÁG 37 of À7.7 kcal/ mol ( Figure 2B ). Thermodynamic stabilities were calculated at the MFOLD website using version 2.3 parameters (http://mfold.rna.albany.edu/?q=mfold/RNA-Folding-Form2.3), as these were previously shown to better fit in vivo hairpin stabilities (29) . These data support the notion that downstream structures serve as barriers to stall translating ribosomes to stimulate frameshifting, and demonstrate that there is a correlation between the thermodynamic stability of a hairpin and its frameshift inducing capacity. A selection of above hairpins was cloned into a dualluciferase reporter plasmid and their frameshifting efficiency assayed in mammalian cells ( Figure 2C) . Although the absolute level of frameshifting was lower than in vitro, the trend was similar and showed maximal frameshifting of $8% around 12-15 bp. The pseudoknot in these assays was 1.6 times more efficient than the 12 and 15 bp hairpins, close to the in vitro ratio of 1.4 (see above). Thus, the hairpin derivative can effectively substitute for the SRV-1 pseudoknot in À1 ribosomal frameshifting. Bulges and mismatches are known to change twisting and bending of a regular stem and are thus expected to influence the way in which a ribosome encounters a hairpin structure (11, 30) . To investigate a possible effect of helical twisting and bending on frameshifting, we introduced mismatches and bulges in the 12 bp stem at a position corresponding to the junction in the SRV pseudoknot ( Figure 3A) . Introduction of an A · A mismatch halfway through the stem (11 bp/AA) decreased frameshifting about 10 fold, although its predicted thermodynamic stability of À25.1 kcal/mol is comparable to that of a regular hairpin of 10 bp, yielding 13% frameshifting ( Figure 2B ). The frameshift inducing ability was recovered when the base pair was restored to A-U (12 bp/AU). We also introduced a single or triple adenosine bulge at either side of the stem, to investigate potential bending effects on frameshifting. Figure 3A and B show that the single adenosine bulge mutant decreased frameshifting, depending on the location of the bulge, five to seven fold compared to the 12 bp hairpin construct. When the bulge was enlarged to three adenosines the frameshifting was almost abolished. Interestingly, the effect of bulges at the 3 0 side of the stem was less dramatic than those at the 5 0 side. The loop composition plays a major role in hairpin stability, RNA/RNA and RNA/protein interaction. These factors may directly influence hairpin-induced ribosomal frameshifting efficiency. To explore the correlation between loop composition and frameshifting efficiency, a number of loop mutations were introduced in the context of a 9 bp stem (Figure 4 ). We note that the UUCG tetraloop with a CG closing base pair (cbp) has higher stability ($2 kcal/mol) than that with a GC cbp (31) . Therefore, we first tested if this different cbp affected frameshifting efficiency. Our results showed that there is no difference in frameshifting efficiency between UUCG and UUCG/cg constructs (Figure 4, bars 6 and 8) . Replacing the UUCG tetraloop by GGGC which, due to its high content of purines, is among the most disfavored tetraloops (32) had only a marginal effect on frameshifting (Figure 4 , compare bars 6 and 7 and Figure 5A, lanes 1 and 4) . Interestingly, increasing the loop size to 9 nt, which is predicted to lower the stability of stem did not affect frameshifting ( Figure 4 , bar 1; Figure 5A , lane 3). Substituting UUCG by another stable tetraloop sequence (GAAA) resulted in a 2-fold decrease in frameshifting ( Figure 5A , lanes 1 and 2) either with GC ( Figure 4 , bar 10) or CG cbp (Figure 4 , bar 9). We designed another five loop mutants to try to explain the low efficiency of the GAAA tetraloop constructs. Constructs AAAA and CAAA induced 5.2% and 4.7% frameshifting, respectively (Figure 4, bars 2 and 3) , which is close to that of the GAAA constructs. The efficiency of two other A-rich loop mutants, ACAA and AAAU, was 7.5% and 7.1%, respectively (Figure 4 , bars 4 and 5), thereby closely matching that of the UUCG constructs. Finally, the GGGA tetraloop construct, belonging to the stable GNRA tetraloop family, induced 1.7-times more frameshifting than its GAAA sibling (Figure 4, bar 11) . These data suggest that the presence of 3 or 4 adenines at the 3 0 side of a tetraloop is unfavorable for frameshifting. To further examine the role of the loop identity or size in ribosomal frameshifting, we cloned some of the above loop mutants into a dual-luciferase reporter plasmid and assayed their frameshifting efficiency in mammalian cells ( Figure 5B ). Our data show that the effects of loop nucleotides are comparable in vitro and in vivo. The stable GAAA tetraloop construct again had the lowest frameshifting efficiency ( Figure 5B , 2.9%), which was half that of the UUCG construct ( Figure 5B, 6 .1%). Most RNA viruses that make use of ribosomal frameshifting employ pseudoknot structures instead of simple hairpins for this job. The reason for this may be the presence of a triple helix interaction between S1 and L2 in most frameshifter pseudoknots, which has been suggested to be a poor substrate for the ribosomal helicase (13, 33) and hence increases ribosomal pausing and the time window for slippage. Although pausing is critical, it is not sufficient for efficient frameshifting (34) . Previously, it was shown that a 17 bp hairpin with a calculated stability of À31.2 kcal/mol derived from the minimal IBV pseudoknot induced 5-to 10-fold less frameshifting in RRL (22) than its parent pseudoknot even though both the hairpin and the pseudoknot can pause ribosomes at the same position and to a similar extent (34) . In the present study, a 12 bp hairpin derivative of the SRV-1 gag-pro pseudoknot with a calculated stability of À26.9 kcal/mol was capable of inducing 22% of frameshifting, which is only 1.4-fold . Influence of loop sequence and closing base pair (cbp) on À1 ribosomal frameshifting efficiency. The composition of various loops capping a 9 bp stem is shown in bold, and CG-cbps are shown in lower case. The constructs are named after their loop sequence followed by the '/cg' extension when the cbp was changed from G-C to C-G. Slippery sequence and spacer are the same as in the construct shown in Figure 1A . Graph is similar to that of Figure 2B except that on the right y-axis ÁG starts from À18 kcal/mol. less than its pseudoknotted counterpart. This indicated that a non-natural hairpin can be an efficient frameshift stimulator, at least in the SRV-1 model. Furthermore, our results showed that the frameshifting efficiency increased upon elongation of the length of the hairpin up to 12-15 bp, which is consistent with our previous data using antisense oligonucleotides of 12-15 nt to induce ribosomal frameshifting (35) . More importantly, the frameshift inducing ability of these hairpin constructs with a perfect stem linearly correlated with the calculated thermodynamic stability, in agreement with two previous reports (19, 20) . In the experiments of Bidou et al. (20) studying the HIV-1 gag-pol frameshift hairpin the stem-length was kept at 11 bp, while its stability was varied between À3.4 and À22.1 kcal/mol (recalculated using MFOLD 2.3) by changing the number of AU and GC base pairs in a small set of six hairpins. In the case of the dnaX gene of Escherichia coli 22 variants of the wild-type 11 bp hairpin were tested for their ability to stimulate À1 PRF at the AAAAAAG slippery sequence. Hairpin stabilities varied between À10.4 and À27 kcal/mol and a positive correlation between frameshifting efficiency and calculated stability was observed both in the presence (R 2 = 0.62) and absence (R 2 = 0.72) of upstream enhancer (19) . The dnaX gene with the highly efficient (prokaryotic) AAAAA AG slippery sequence is not directly comparable to our in vitro system; a 6 bp hairpin in the dnaX gene displayed 17% of frameshifting without upstream enhancer, whereas a 6 bp hairpin in our system induced only 3.5% of frameshifting. In the HIV-1 gag-pol gene Bidou et al. (20) observed a 15-20% decrease in frameshifting in vivo with their most stable hairpin, similar to our results with the 21 bp hairpin. However, in our case, the stability at which this happened was À45 kcal/mol much higher than their most stable hairpin of À22.1 kcal/mol. It is possible that this difference is due to the different experimental systems. Although it has been suggested that too stable stems increase the time for tRNAs to shift back into the 0-frame again (20) we believe that our 21 bp hairpin is less efficient because it has more AU bps in the middle of the stem compared to the 12 and 15 bp hairpins (Figure 2A) . The experiments with hairpins harboring bulges or mismatches halfway through the stem demonstrated that this region is quite important for frameshifting ( Figure 3A and B) . Even though the overall stability of these constructs was comparable to that of a hairpin of 9 or 10 bp, their frameshift activity was equal or lower than that of a 6 bp hairpin of À13.1 kcal/mol: as if the mismatch or bulge after the 6th base pair disconnected the upper part of the stem. This observation is reminiscent of the overall destabilizing effect of mismatches in DNA hairpins. In a pioneer singlemolecule pulling study, it was shown that introducing a mismatch in a 20 bp DNA hairpin shifted its transition state close to the location of the mismatch (36) . Our data also comply with this mechanical study and suggest that mechanical stability may be a better parameter than thermodynamic stability to describe the frameshift efficiency of hairpins. In addition to the mentioned dnaX and HIV-1 gag-pol hairpins, other examples of frameshifter hairpins are found in HTLV-2 and CfMV ( Figure 6 ). HTLV-2 gag-pro features a perfect 10 bp hairpin with CUA tri-loop which induces 9% frameshifting in RRL (16) . The CfMV 2a-2b frameshifting hairpin consists of 12 bp, one cytidine bulge close to the top, and a stable UACG tetraloop and is capable of inducing 11% of frameshifting in a wheat germ cell-free system (WGE) (17) . What these hairpins have in common is their length of 10-12 bp, their relatively low number of mismatches and bulges, their small loops and their high GC content, especially in the bottom 6 bp. These features are also applicable to the good frameshifters from our dataset. Interestingly, these features do not all apply to the minimal IBV hairpin ( Figure 6 ) that is derived from the so-called minimal IBV pseudoknot. Despite its large size of 17 bp, absence of mismatches and bulges, presence of a small loop, the stability of the middle part of the hairpin, i.e. bp 5-9, is not very high. This could be the reason why its activity in RRL is 5-10 fold lower (22) than of its parent pseudoknot, whose activity is 42% (25) . Surprisingly, in our assays the frameshift-inducing efficiency of the IBV hairpin was 26% (data not shown), which is in stark contrast to the 4-8% reported by Brierley et al. (22) . This discrepancy may be due to experimental conditions: in our experiments we used non-capped transcripts, a 7-nt spacer and RRL from Promega whereas the Brierley's lab used capped transcripts, a 6-nt spacer and in-house prepared RRL. On the other hand, the 26% we obtained for the IBV hairpin would be a factor of 1.6 lower than the 42% reported for the IBV pseudoknot (25) , and is similar to the ratio of 1.4 and 1.6 we obtained for SRV in vitro and in vivo, respectively. Remarkably, in WGE the IBV hairpin has been reported to induce high levels (34%) of frameshifting versus 51% for the IBV pseudoknot (24) . In that study modified extracts were used that are somewhat more frameshift-prone than the standard wheat-germ extracts. Nevertheless, the ratio between pseudoknot and hairpin-induced frameshifting in this system is also 1.5. This number may reflect the additional interactions, like base triples, in a pseudoknot that make it a better frameshift stimulator than a hairpin. In addition to stem size, loop composition is another determinant of hairpin stability. An important subgroup of hairpin loops is the tetraloop, which is the most common loop size in 16S and 23S ribosomal RNAs (37) . The tetraloops with consensus UNCG, GNRA, or CUUG loop sequence form stable loop conformations (38, 39) . As opposed to the mentioned stable tetraloops, purine-rich (32) and larger loops (40) are considered to be less favorable for hairpin formation. Our results showed that the GGGC loop is indeed less efficient in inducing frameshifting but the larger loop construct (9 bp/9 nt), although having a lower thermodynamic stability, showed comparable frameshifting efficiency to the stable UUCG tetraloop hairpin. This is consistent with previous studies that showed that increasing the size of the loop in a hairpin or pseudoknot can increase frameshift-inducing ability to a certain extent (21, 41) . Although larger loops seem efficient in inducing frameshifting, in known examples of frameshifter hairpins, there are no loop sizes of more than 5 nt. This could relate to hairpin folding kinetics (40) or to nuclease sensitivity. Intriguingly, we found that a 9 bp stem capped with a GAAA tetraloop is 2-fold less efficient in inducing frameshifting than its UUCG counterpart in vitro and in vivo. It has been reported that GAAA tetraloops are frequently involved in RNA tertiary interactions (42) . We hypothesize that the GAAA tetraloop may be involved in an unknown RNA tertiary structure with ribosomal RNA, thereby interfering with frameshifting. The fact that in the known natural examples of frameshifter hairpins, the GAAA tetraloop, despite its high stability, is absent can be taken as support for this hypothesis (Olsthoorn, unpublished data) . Further investigation of this observation may lead to new insights in ribosomal frameshifting. In conclusion, our data show that hairpins of various base composition in stem and loop can act as efficient frameshift stimulators. Combined with previous studies on antisense-induced frameshifting (43, 44) , these data support the notion that downstream structures primarily serve as barriers to stall translating ribosomes to stimulate frameshifting. Although there exists a linear relationship between calculated stability and frameshifting, local destabilizing elements like bulges or mismatches in a hairpin can greatly influence frameshift-inducing activity. Future experiments addressing the mechanical strength of these hairpins (7-9) may help to improve our understanding of the basics of ribosomal frameshifting.

Usefulness of Published PCR Primers in Detecting Human Rhinovirus Infection

We conducted a preliminary comparison of the relative sensitivity of a cross-section of published human rhinovirus (HRV)–specific PCR primer pairs, varying the oligonucleotides and annealing temperature. None of the pairs could detect all HRVs in 2 panels of genotyped clinical specimens; >1 PCR is required for accurate description of HRV epidemiology.

H uman rhinoviruses (HRVs) cause more asthma exacerbations than any other known factor, in addition to causing most colds and infl uenza-like illnesses. The prevalence of HRV in published reports varies considerably. A novel HRV clade identifi ed in 2006, now known as HRV species C (HRV-C) (1), can be identifi ed only by PCR. Since 1988, seasonality and clinical outcomes and numerous different primer pairs have been used to identify HRV; how well these methods perform on new HRV types is uncertain. Given the likely variation in the preparation of RNA, the quality and formulations of commercial reverse transcription (RT)-PCR enzymes and reaction mix components and changes in thermal cyclers since 1988, not surprisingly many, perhaps most, of these assays are not being used in the manner they were originally described. For example, the fi rst HRV-specifi c primers reported (2) have subsequently been used with different RNA preparation methods, amounts of reverse transcriptase, cDNA priming strategies, dNTP concentrations, annealing temperatures (T M s), and cycling conditions (3, 4) . We conducted a preliminary comparison of the relative sensitivity of a cross-section of published HRV-specifi c PCR primer pairs (most of which were fi rst published before HRV-C was reported), independent of most variables described above, by testing a panel of 57 clinical specimen nucleic acid extracts from combined nose and throat swabs from preschool children with colds and infl uenza-like illnesses in Melbourne, Australia. The study was approved by the Royal Children's Hospital Human Research Ethics Committee. The panel included representatives of the 3 HRV species (Figure) , human enteroviruses (HEVs), and extracts negative for picornaviruses. The HRVs had been previously detected by using a nested primer pair (online Appendix Table, www.cdc.gov/EID/content/17/2/294-appT.htm) (5). We used 10 different HRV primer pairs and also retested specimens by using the original primer pair with our standard reagents and equipment (5) . We applied the published T M when possible. The original descriptions of primer pairs 7 and 10 (online Appendix Table) lacked T M information, and after in-house calculations, we used T M s of 50°C and 58°C, respectively. We also deliberately standardized the reagents (OneStep RT-PCR kit, QIA-GEN, Doncaster, Victoria, Australia) and thermal cyclers used (Veriti, Applied Biosystems, Foster City, CA, USA) for conventional PCR and the RotorGene 3000 real-time cycler (QIAGEN). Because primer pair 1 had a published history of detecting types from all HRV species, we chose it to genotype HRV-positive samples by sequencing the amplifi ed products. Other pairs were used if pair 1 was unsuccessful. We found that no primer pair detected the same HRVs and HEVs typed when the original pair (5) or pair 1 (online Appendix Table) was used. Five primer pairs, including real-time PCR (rtPCR) pair 5, did not amplify the HEVs, a positive feature for HRV-specifi c studies. Only 2 primer pairs amplifi ed anything from a specimen that was positive for both HRV and HEV, a problem for accurate estimation of the frequency of co-detections. The original primer pair screen detected 3 untypeable picornaviruses, which were not detected by any other pair or by repeat testing using the same pair. Only the second-round amplicon of the 3 nested sets of nested primer pairs (2, 3, and 9) was considered because the second round increased the total number of positive specimens over the fi rst round. The longest amplicon, produced by primer pair 7, was also a valuable genotyping target, but it detected only 14 of the original 27 HRV-positive specimens in this population. We next selected 4 frequently published primer pairs (1, 5, 7, and 8) to examine 44 picornavirus-positive specimens (39 HRVs, 3 HEVs, and 2 untypeable picornaviruses) from nonhospitalized children with acute asthma exacerbation (6) . As before, primer pair 1 detected the greatest num- Most notably, primer pair 7 performed better than it had in the previous population, detecting only 1 fewer HRV than primer pair 1 and 9 more HRVs than pair 8. No speciesspecifi c bias was apparent, but generally, a specimen with a lower RNA concentration, as indicated by the cycle threshold from primer pair 5, was less likely to be detected or typed by using other primer pairs. Primer pairs 5 and 8 did not detect the 3 HEVs (HEV-68). We noted in both populations that primer pair 1 sometimes amplifi ed a region of human genomic DNA from chromosome 6 (GQ497714), for which amplicon size was indistinguishable from that expected due to HRV. It was not possible to use the precise conditions reported for the 10 compared assays; 1 was published >2 decades ago and used phenol chloroform extraction. Some of the original enzyme formulations or reagents are no longer available, and production processes have changed in the interim. Thermal cyclers have also changed. There was no consensus on enzymes and reaction mixes used. In addi-tion, the previously published primers were used in assays divided between those using 1-step RT-PCR and those using a separate RT cDNA synthesis step. A review of studies that detected HRVs with adequately described conditions during 2009-2010 found that fewer used a single-tube RT-PCR approach than a 2-step system. We conducted singletube RT-PCR to maintain the benefi ts of the so-called closed amplifi cation system of rtPCR. Thus, we chose to use a single common set of reagents as the fairest way to compare the primer pairs examined in this study. We believe the nature of this relative comparison best refl ects performance for the likely end users: clinical microbiology laboratories or researchers. We compared primers rather than assay function using clinical material instead of cultured virus, plasmid or synthetic RNA standards, or screening contemporary or archived extracts, which are sometimes of low viral load. When picornavirus epidemiology is the primary research focus, we recommend using >2 primer pairs to maximize the detection of HRVs. Under our conditions, pairs 1-4 returned the highest number of positive results, and the rtP-CRs behaved similarly but with reduced sensitivity. The rtPCR that used pair 5 did not amplify known HEVs. Many possible reasons could cause discrepant virus testing results between different sites, including changes to specimen integrity resulting from transport and variable amplifi cation resulting from low viral loads. The effects of viral load can be seen in this study: specimens in population 1 that were positive with multiple (>6 separate pairs) primer pairs had a mean cycle threshold of 33.3 (combining results from both rtPCRs), whereas those with <6 positive results had means of 39.3 cycles. Most (29/33) specimens with <3 positive primer pairs were negative by rtPCR. Amplifi cation variability can also be attributed to the substantial nucleotide sequence diversity between HRVs and the different temporal and clinical characteristics of the 2 specimen populations we used. Population diversity is a feature of HRV studies in the literature. Our selection of published primer pairs includes those from studies that have informed our current understanding of HRV epidemiology. Finding such a high degree of variability in performance was thus noteworthy. Ineffi cient HRV detection by PCR may be a serious problem for research studies. Comparison of data between different HRV studies is confounded as are data from studies seeking to determine the effects of other respiratory viruses. The prevalence, seasonality, transmission, and clinical effects of HRV types and species require reexamination with tools that have been comparatively validated to ensure their sensitivity. Figure. Distribution of human rhinovirus (HRV) and human enterovirus (HEV) sequences used for primer pair studies. The HRV and HEV genotypes from the testing panel (indicated by fi lled circles) were aligned with the central 154 nt of the 5′ untranslated region (UTR) region of all complete HRV genomes and poliovirus-1. HRV-Ca and HRV-Cc refer to HRV-Cs with 5′ UTR sequences that have phylogenetic origins from either HRV-As or HRV-Cs, respectively. The tree was constructed by neighbor joining of maximum composite likelihood distance implemented in MEGA (www.megasoftware.net).

NNAlign: A Web-Based Prediction Method Allowing Non-Expert End-User Discovery of Sequence Motifs in Quantitative Peptide Data

Recent advances in high-throughput technologies have made it possible to generate both gene and protein sequence data at an unprecedented rate and scale thereby enabling entirely new “omics”-based approaches towards the analysis of complex biological processes. However, the amount and complexity of data that even a single experiment can produce seriously challenges researchers with limited bioinformatics expertise, who need to handle, analyze and interpret the data before it can be understood in a biological context. Thus, there is an unmet need for tools allowing non-bioinformatics users to interpret large data sets. We have recently developed a method, NNAlign, which is generally applicable to any biological problem where quantitative peptide data is available. This method efficiently identifies underlying sequence patterns by simultaneously aligning peptide sequences and identifying motifs associated with quantitative readouts. Here, we provide a web-based implementation of NNAlign allowing non-expert end-users to submit their data (optionally adjusting method parameters), and in return receive a trained method (including a visual representation of the identified motif) that subsequently can be used as prediction method and applied to unknown proteins/peptides. We have successfully applied this method to several different data sets including peptide microarray-derived sets containing more than 100,000 data points. NNAlign is available online at http://www.cbs.dtu.dk/services/NNAlign.

Proteins are extremely variable, flexible and pliable building blocks of life that are crucially involved in almost all biological processes. Many diseases are caused by protein aberrations, and proteins are frequent targets of intervention. A plethora of highthroughput methods are currently being used to study genetic associations and protein interactions, and intense on-going international efforts aim at understanding the structures, functions and molecular interactions of all proteins of organisms of interest (e.g. the Human Proteome Project, HPP). In some cases, linear peptides can emulate functional and/or structural aspects of a target structure. Such peptides are currently identified using simple peptide libraries of a few hundreds to thousands peptides whose sequences have been systematically derived from the target structure at hand -that is, if this is known. Even when the native target structure is unknown, or too complex (e.g. discontinuous) to be represented by homologous peptides, the enormous diversity and plasticity of peptides may allow one or more peptides to mimic relevant aspects of a given target structure [1, 2] . Peptides are therefore of considerable biological interest and so are methods aimed at identifying and understanding peptide sequence motifs associated with biological processes in health and disease. Indeed, recent developments in large-scale, high-density peptide microarray technologies allow the parallel detection of thousands of sequences in a single experiment, and have been used in a wide range of applications, including antibody-antigen interactions, peptide-MHC interactions, substrate profiling, identification of modification sites (e.g. phosphorylation sites), and other peptide-ligand interactions [3, 4, 5, 6, 7] . One of the major advances of peptide microarrays is the ease of generating large numbers of potential target structures and systematic variants hereof [8] . Given the capability for large-scale data-generation already realized in current ''omics'' and peptide microarray-based approaches, experimentalists will increasingly be confronted with extraordinary large data sets and the consequent problem of identifying and characterizing features common to subsets of the data. These are by no means trivial problems. Up to a certain level of size and complexity, data can be presented in simple tabular forms or in charts, however, larger and/or more complex bodies of data (e.g. in proteome databases) will need to be fed into bioinformatics data mining systems that can be used for automated interpretation and validation of the results, and eventually for in silico mapping of peptide targets. Moreover, such systems can conveniently be used to design next-generation experiments aimed at extending the description of target structures identified in previous analyses [9] . A wealth of methods has been developed to interpret quantitative peptide sequence data representing specific biological problems. By way of examples, SignalP, which identifies the presence of signal peptidase I cleavage sites, is a popular method for the prediction of signal peptides [10] ; LipoP, which identifies peptidase II cleavage sites, predicts lipoprotein signal peptides in Gram-negative bacteria [11] ; various prediction methods predict phosphorylation sites by identifying short amino acid sequence motifs surrounding a suitable acceptor residue [12, 13, 14, 15] etc. In general terms, these methods can be divided in two major groups depending on the structural properties of the biological receptor investigated, and of the nature of the peptides recognized. The simplest situation deals with interactions where a receptor binds peptides that are in register and of a known length. In this case, the peptide data is pre-aligned, and conventional fixed length, alignment-free pattern recognition methods like position specific weight matrices (PSSM), artificial neural networks (ANN), and support vector machines (SVM) can be used. Peptide-MHC class I binding is a prominent example of the successful use of such methods to characterize receptor-ligand interaction represented by pre-aligned data (reviewed in [16] ). Another more complex type of problems deals with interactions where either the motif lengths, and/or the binding registers, are unknown. In these cases, the peptide data must a priori be assumed to be unaligned and any bioinformatics method dealing with such data is faced with the challenge of simultaneously recognizing the binding register (i.e. performing an alignment) and identifying the binding motif (i.e. performing a specificity analysis). Peptide-MHC class II binding is a preeminent example of a receptor-ligand interaction represented by unaligned data. Several bioinformatics methods have been developed to identify binding motifs in such peptide data including Gibbs sampling [17] , hidden Markov models (HMM) [18] , stabilization matrix method (SMM) alignment [19] , and alignment using artificial neural networks [20] (for more references see [21] ). Another example of unaligned peptide data is that of antibodies interacting with linear peptide epitopes. Although B-cell epitopes frequently are conformational and three-dimensional in structure, some do contain linear components that can be represented by peptide interaction with the corresponding antibodies [22, 23, 24] . Even though most of the methods described above are standard methods for data-driven pattern recognition, the development of a prediction method for any given biological problem is far from straightforward, and the non-expert user will rarely be able to develop their own state-of-the-art prediction methods. We have recently described a neural network-based data driven method, NN-align, which has been specifically designed to automatically capture motifs hidden in unaligned peptide data [20] . NN-align is implemented as a conventional feed-forward neural network and consists of a two-step procedure that simultaneously identifies the optimal peptide-binding core, and the optimal configuration of the network weights (i.e. the motif). This method is therefore inherently designed to deal with unaligned peptide data, and it identifies a core of consecutive amino acids within the peptide sequences that constitute an informative motif. Note that the method does not allow for gaps in the alignment. Although NNalign was originally developed with the unaligned nature of peptide-MHC class II interaction in mind -and independent validations have shown that NN-align indeed performs significantly better than any previously published methods for MHC class II motif recognition [25] -the unique ability of this method to capture subtle linear sequence motifs in quantitative peptide-based data and its adaptability makes it extremely attractive for other applications as well. Here, we have adapted and extended the NNalign method so that it can handle quantitative peptide-based data in general. Making this method generally available for the scientific community, we have embedded it into a public online web-interface that facilitates both handling of input data, optimization of essential training parameters, visual interpretation of the results, and the option of using the resulting method to predict on user-specified proteins/peptides. Through the server the user can easily set up a cross-validation experiment to estimate the predictive performance of the trained method, and automatically reduce redundancy in the data. The logo visualization is also improved with an algorithm that aligns individual neural networks to maximize the information content of the combined alignment. This web-based extension of the NN-align method empowers experimentalists of limited bioinformatics background with the ability to perform advanced bioinformatics-driven analysis of his/ her own sets of large-scale data. Enabling any non-expert end-user to extract specific information from quantitative peptide data using an advanced bioinformatics approach, we have used our recently published NN-align method to generate a web-based extension with a reasonably simple, yet adaptable, web-interface and made this server publicly available at http://www.cbs.dtu.dk/services/NNAlign. Using this web server any user can submit quantitative peptide data (optimally based on actual discrete measurements, but even assigned classification, e.g. 0 and 1, can be used) and in return receive a trained method including training details and estimated predictive performance, a visual interpretation of the identified peptide pattern, and the trained model itself. The latter can be resubmitted to the web server at any later time and used to predict the occurrence of the learned motif in one or more concurrently submitted peptide sequences or FASTA format sequences. The truly non-expert user has the option of using a set of default settings. Using these settings, the data is preprocessed using a linear transformation to make the data fall in the range from 0 to 1, and the NN-align method is trained using five-fold crossvalidation. For each cross validation partition five networks, each initiated from different initial configurations, are trained with 3 hidden neurons. The only critical parameter that the user is required to specify is the motif length. The value used for this parameter is specific to each problem and the user is recommended to define a motif length (or an interval of motif length) that is relevant to the biological problem investigated by the peptide data. The default settings will in most cases allow the user to obtain a first impression of the motif contained in the data, and achieve a prediction method that allows the user to make prospective studies on uncharacterized proteins/peptides. The more experienced user has several advanced options to customize the training. For details on these options refer to Materials and Methods section, or the help section of the web-server. An example output from the NNAlign Server is shown in Figure 1 . Information about the training data is accompanied by a plot of the data distribution before and after the data processing needed to train the neural networks. An important feature is the possibility to download and save the trained model, and use it subsequently for predictions on new data. The results page also returns the performance of the method as estimated by crossvalidation, and provides links to a scatter-plot showing the correlation between measured and predicted values, as well as the complete alignment core on the training data. A sequence logo gives a visual representation of the identified sequence motif, which can also be viewed in a log-odds position-specific scoring matrix format. If any evaluation data has been provided at the time of method training, a section of the results will report the predictions of this evaluation set. A few example applications illustrating the power of the NNAlign method are presented in the following sections. First, the method is applied to examples of pre-aligned peptide data using examples of MHC class I binding. Next, the alignment problem is included using MHC class II binding data, showing the ability of the method to identify at the same time the correct length of the motif, the binding register, and the sequence motif itself. An important output from the NNAlign method is a sequence logo representing the identified binding motif. Such sequence logos provide a highly intuitive representation of single-receptor specificities (as is the case Figure 1 . Example of output from the NNAlign server trained on MHC class II binding data for allele HLA-DRB1*0101. Links on the results page (in pink) redirect to additional files and figures relevant for the analysis. Run ID is a sequential identifier for the current job, and Run Name a user-defined prefix that is added to all files of the run. The ''view data distribution'' link shows the transformation applied to the data in preprocessing, which can be either a linear or logarithmic transformation. In this case the method was trained with a motif length of 9, including a PFR of size 3 to both ends of the peptide, and encoding in the network input layer peptide length and PFR length. The hidden layer was made of a fixed number of 20 neurons. Peptides were presented to the networks using a Blosum encoding to account for amino acid similarity, for 500 hundred iterations per peptide without stopping on the best test set performance. At each cross-validation step, 10 networks were trained starting from 10 different initial configurations. The subsets for cross-validation were constructed using a Hobohm1 method that groups in the same subset sequences that align with more than 80% identity (thr = 0.8). The model can be downloaded to disk using the dedicated link, and can be resubmitted to NNAlign to find occurrences of the learned pattern in new data. The estimated performance of the trained method is expressed in terms of Root Mean Square Error, Pearson and Spearman correlation. A visual representation of the correlation can be obtained from the scatterplot of predicted versus observed values. The ''complete alignment core'' link allows downloading the prediction values in cross-validation for each peptide, and where the core was placed within the peptides. Next follows a section on the sequence logo, showing a logo representation of the binding motif learned by the network ensemble. If the relative option is selected, links to logos for the individual networks in the final ensemble are also listed here. Finally, if an evaluation set is uploaded, an additional section shows performance measures and core alignment for these data. doi:10.1371/journal.pone.0026781.g001 for MHC class I and II binding data). Finally, to illustrate how the method is capable of handling and guide the semi-expert user in interpreting large-scale data sets, NNAlign is applied to data generated by a large-scale peptide microarray technology. Binding of peptides to MHC class I molecules is highly specific, with only 1-5% of a set of random natural peptides binding to any given MHC molecule [26] . Moreover, in the vast majority of cases only peptides with length 8-10 amino acids can fit in the binding pocket of MHC class I molecules. The predictive performance of NNAlign on 12 human MHC class I alleles from data by Peters et al. [27] is shown in Table 1 (see the table footnote for the parameters used). The benchmark data sets contain quantitative binding data of a given length (9 amino acids) covering the whole spectrum from non-binding to strong-binding peptides, hence serving as a perfect illustration of the strength of the NNAlign method to handle pre-aligned peptide data. The overall performances of the three methods are comparable demonstrating that NNAlign competes with state-of-the-art methods designed specifically for MHC class I prediction. As opposed to MHC class I binding, which is mostly limited to peptides of similar length, the MHC class II molecule interacts with peptides of a wide length distribution and high compositional diversity [28] . Binding of a peptide to an MHC class II molecule is primarily determined by a core of normally 9 amino acids, but the composition of the regions flanking the binding core (peptide flanking region, PFR) has been shown to also affect the binding strength of a peptide [29, 30] . Identifying the binding motif and binding register for MHC class II binding peptides is thus a problem that inherently requires simultaneous alignment and binding affinity identification. Here, an MHC class II benchmarking was obtained from the recent publication by Wang et al. [25] . The performance was estimated for each allele using a 5 fold cross validation, where at each step 4/5 of the data were used to train the neural networks, and 1/5 were left out for evaluation. For cross-validation, we preserved the same data partitioning as used in the original publication. In Table 1 , the performance of NNAlign on the Wang set is compared to other publicly available methods for MHC class II prediction. These include SMM-align [19] , ProPred/Tepitope [31, 32] , as well as the original version of the NN-align algorithm [20] . The NN-align-based methods outperform their competitors on all alleles, confirming the ability of the neural networks in dealing with alignment problems. The difference with the original NN-align method, which is due to differences in network architecture, is small and not significant (p.0.2, binomial test). For this example involving unaligned data, the NNAlign server competes with comparable state-of-the-art methods. Choosing the optimal motif length Different positions in a binding motif can be more or less informative, and the ends of a motif can often not be clearly delineated. This prompts the question of how many positions are necessary and sufficient to represent a given motif and how the length of a motif is defined. NNAlign allows searching for the optimal motif length in a quantitative peptide data set. Here, the best motif length is the one that yields, in a cross-validation experiment, the lowest root mean square error (RMSE) between For MHC class I no significant difference is found in predicted performance between the NNAlign, SMM and ANN method (p.0.5, binomial test). The values for the SMM and ANN methods were taken from Peters et al. [27] . The method was trained using a fixed motif length of 9 corresponding to the peptide length, and constructing a network ensemble with multiple architectures using respectively 2,3,4,5 and 7 hidden neurons. Performance was measured in cross-validation, training each network for a fixed number of 500 iterations per sequence. The different MHC class II prediction methods are NN-align [20] , SMM-align [19] , and Propred [31, 32] . NNAlign server is the method described here. Performance values for first 4 methods are taken from [25] . NNAlign was trained with a motif length of 9, flanking regions of 3 amino acids, Blosum encoding including peptide length and flanking region length, and an ensemble of 2, 3, 5, 9 and 12 hidden neurons for each of 10 initial random configurations. In bold is highlighted the best performing method for each MHC allele. The column # gives the number of the peptides in the data set for the given allele. doi:10.1371/journal.pone.0026781.t001 observed and predicted values. By this token, a terminal position is included in the motif if it contributes with information at a level above what could be considered to be noise. In contrast, if the inclusion of a putative terminal position does not lead to a reduction in the RSME then it can be concluded that it does not add useful parameters to the model; rather, it lowers the predictive performance and should be omitted. This approach was used to suggest the motif length of the 14 MHC class II HLA-DR alleles, which were searched for optimal predictive performance by scanning through possible lengths from 6 to 11 amino acids. NNalign will report the length associated with the lowest RMSE value as the optimal motif length (see Figure 2 , left hand panel). Nonetheless, the user is advised to inspect the sequence logo as well as the performance plot of the RMSE as a function of the motif length to evaluate whether the dependence upon length appears significant. As defined here and illustrated in Figure 2 right panel, the 9-mer preference of HLA-DRB1*01:01 is significant, whereas the apparent 8-mer preference of HLA-DRB1*15:01 is not significant. In fact, for the 14 HLA-DR molecules included in the benchmark, only one was found to have a single consistent optimal motif length (DRB1*0101 with a motif length of 9 amino acids). For all other molecules the method did identify more than one possible optimal motif length. However, all motif lengths fell in the range of 7 to 10 amino acids, and in all cases a 9-mer motif was compatible with being the optimal motif length. In order to enhance predictive performance, the NNAlign method exploits an ensemble of neural networks [20, 33] , which have been trained on different subsets of the data, and/or from alternative configurations of the network architecture (i.e. different number of hidden neurons and/or encoding schemes). As a consequence of different architectures and starting conditions, individual networks might disagree on the exact boundaries of the motif. This disagreement would complicate the visualization of the motif if this was represented as a simple overlay of the individual motifs as exemplified in Figure 3 , where sequence logos for four different networks from the ensemble trained on HLA*DRB1-04:01 binding data are shown in panels A through D. The individual networks agree on identifying the same strong primary anchor residues and positions, however, each single network identifies different ends (i.e. suggests different registers of the same motif; in casu starting at positions 1, 2, 2 and 3 of the predicted nonamer peptide). The weak C-terminal primary anchor residue of HLA*DRB1-04:01 probably explains why the boundaries are difficult to determine. A simple overlay of the predictions from individual networks would result in a muddled motif as depicted in Figure 3 , panel E. Implementing a Gibbs sampler approach, where matrix representations of the core motifs of different networks are aligned, we introduced an off-set correction for each network aiming at maximizing the information content of a combined logo representation of the motif. This approach led to a considerable improvement in the visual logo representation of the binding motif (Figure 3, panel F) . Offset correction is included as an integral part of the method to enhance motif visualization. Characterizing the binding motif of HLA-DR molecules using the NNAlign method To illustrate the power of the NNAlign method to capture the binding motifs within unaligned quantitative peptide data, we applied the method to derive sequence logo representations of the 14 MHC class II HLA-DR molecules included the Wang dataset. NNAlign was trained with a binding motif length of 9 amino acids, Blosum encoding, including peptide length and flanking region length, and PFRs of 3 amino acids, homology clustering at threshold 0.8 using all data points, 20 hidden neurons and a 5-fold cross-validation without stopping at the best test set performance. These parameters were found to be optimal in the original NNalign paper for MHC class II binding prediction [20] , with the only difference that here we choose a single value for hidden layer size for a matter of prediction speed. Individual networks are aligned to a common register using the offset correction strategy previously described. The sequence logos obtained are shown in Figure 4 . The sequence logos reflect the overall consensus of the binding motifs for HLA-DR molecules, namely a prominent P1 anchor with strong amino acids preference towards hydrophobic Figure 2 . Identification of optimal motif length using the NNAlign method. Left panel: Histogram of the optimal motifs lengths for the 14 HLA-DR molecules in the Wang dataset as identified by the NNAlign method. Right panel: Predictive performance measured in terms of the root mean square error (RMSE) between observed and predicted values as a function of the motif length for the two molecules DRB1*0101 and DRB1*1501. NNAlign was trained using the same parameters settings described in Figure 4 . At each motif length are shown the mean and standard error of the mean RMSE as estimated by bootstrap sampling. For DRB1*0101 a single consistent optimal motif length of 9 amino acids is found. For DRB1*1501 all motif length 8-11 had statistically indistinguishable performance (paired t-test). doi:10.1371/journal.pone.0026781.g002 amino acids in general, and aromatic amino acids as F and Y in particular, and the presence of two or more additional anchors at P4, P6 and/or P9 each with a unique amino acid preference. Even though most of these motifs exhibit a strong preference for hydrophobic and neutral amino acids at most anchor positions, some dramatic deviations from this general pattern exist. Examples of this are the motifs of DRB1*0301 and DRB1*1101 molecules that have strong preferences for charged amino acids at P4 and P6, respectively. Handling large data sets exemplified by protease recognition of high-density peptide microarrays A peptide microarray containing a total of .100,000 peptides (49,838 of which were unique) was digested with the protease trypsin. The peptide sequences had been synthesized using the theme Ac-GAGAXXXXXGAGA, where Ac-is acetyl blocking the peptide alpha-amino group prior to digestion, and X represents amino acids chosen randomly from the 20 natural amino acids (except lysine, as this residue contains an epsilon-amino group, which even without digestion would be detectable (see Materials and Methods for details)). As a result, free amino groups can only be expressed by trypsin cleaved peptides, which then can then be labeled with Dylight549 and quantitated by fluorescence microscopy. A fluorescence microscopy picture of such a digested and stained peptide microarray (Figure 5a ) demonstrates both the resolution of the photolithographic peptide synthesis strategy and the dynamic range of the free amino group detection strategy. The resulting data was log-transformed and rescaled to obtain a data distribution covering the spectrum between 0 and 1, which -along with the corresponding peptide sequences encoded as Blosum scores without flanking regionswere used to train an NNAlign method. Training was done with a motif length of 5, a fixed number of 3 hidden neurons, 5-fold exhaustive validation, and stopping at the best test set performance. The prediction method yielded a Pearson correlation between measured values and predictions of r = 0.971, a Spearman correlation of r = 0.910, and receiver operating characteristics (ROC) area under the curve (AUC) of 0.997 (using The fundamental pattern appears in all these networks, but they place the anchors at different position of the core. e) shows the core of the 20 networks ensemble without offset correction; in f) offset correction was used to realign the logos to a common register. doi:10.1371/journal.pone.0026781.g003 a target threshold of t = 0.5). The very high performance measures of the resulting NNAlign method demonstrate both that the recorded peptide digestion data contains a consistent and intelligible signal, and that the NNAlign method is capable of deciphering and predicting this extraordinary large number of sequence-dependent peptide signals. The correlation scatterplot feature of the NNAlign web-server output, which compares predicted vs. observed values, further supports the validity of both the peptide microarray and of the NNAlign method. The correlation scatterplot for the trypsin digestion data reveals two major populations of peptides, one composed of non-degradable, non-predicted peptides and one containing weakly to strongly degradable, predicted peptides ( Figure 5b) . Few (0.7%) of the former peptides contained Arginine, whereas most (97.1%) of latter peptides contained Arginine. This is exactly what one would have expected from a peptide digestion with trypsin, which is known to cleave at the C-terminal side of amino acids Arginine (and Lysine, which has been excluded here, see above) [34] . For illustration purposes, Figure 5b includes a color-enhanced visualization of certain dipeptide sequences (note, this is not a standard feature of the NNAlign server) showing that RP sequences are resistant, RA sequences are quite susceptible, and RR sequences appear extremely susceptible to trypsin digestion. Thus, the known trypsin resistance of RP sequences is both demonstrated by the peptide microarray and subsequently captured by the NNAlign method. Note, that both the peptide microarray and the NNAlign generate a continuous set of measurements and predictions showing that trypsin cleavage involves a more complex interaction than a simple recognition solely of an Arginine residue (and by inference a Lysine residue), which would have resulted in a cleaved/non-cleaved classification [35] . It is also important to note that the detection strategy employed here does not reveal where the protease cleavage has occurred, but merely that the protease has recognized the peptide as a substrate and cleaved it somewhere. A similar high-density peptide microarray driven approach was next used to address the specificity of the protease chymotrypsin, which is known to preferentially cleave at the C-terminal of tyrosine, phenylalanine and tryptophan (albeit not if followed by a proline). A high-density peptide microarray containing about 50,000 peptides (16,526 unique peptides) was generated according to the theme Ac-GAGAXXXXGAGA, treated with chymotrypsin, labeled with TAMRA and quantitated by fluorescence microscopy. The resulting data was used to train an NNAlign method (using the settings described in Figure 5 ). The correlation scatterplot of the measured versus predicted values exhibits a very strong linear correlation with a Pearson of r = 0.943 demonstrating that the peptide microarray data contains a consistent signal that reliably has been captured by the NNAlign method. The amount of data deposited in genomic and proteomic databases has been growing exponentially for many years [36] . Due to recent technological advances that have enabled wholegenome sequencing and made whole-proteome analysis a realistic goal, sequence data will accumulate at an even faster pace in the future where single laboratories, even single experiments, can generate data at the ''omics'' level. This is amply illustrated here where a high-density peptide microarray technology allowed the parallel synthesis of more than 100,000 discrete peptide sequences per array, and the collection of a corresponding number of quantitative peptide-receptor interaction data -all within a single experiment. The biggest hurdle of future ''omics'' research may easily become that of making sense of such large-scale biologic sequence data [37] . Presently, the ''omics'' experimentalist requires assistance from specialized and highly trained bioinformaticians capable of large-scale data handling and interpretation. Ideally, however, he or she should not only be armed with highthroughput data-generation technologies, but also with reasonably easy and robust bioinformatics methods allowing the experimentalist to analyze his or her own data. This would permit an immediate analysis of experimental results and assist in rational designs of next generation experiments aimed at extending the original analysis e.g. providing in silico tools for searches that potentially could encompass entire proteomes. Enabling the same person to do large-scale experiments and analysis should result in a better integration between design, experiment, and interpretation -and eventually support the development of new hypotheses. Unfortunately, suitable bioinformatics resources aimed at the nonexpert user are currently scarce, and rarely web-based. In our experience, open source software packages such as Weka [38] are not capable of performing concurrent alignment and motif identification, and are not suited for treating large-scale data sets. A widely used method for motif discovery, MEME [39] , can perform searches for un-gapped sequence patterns in DNA or protein sequences, and offers a user-friendly online server to the untrained user. However, this method is not designed for use in quantitative data, such as peptide-MHC binding or peptide microarray data. To the best of our knowledge, NNAlign is the first web-based bioinformatics solution that allows non-expert users to discover short sequence motifs in quantitative peptide data. As shown here, NNAlign easily competes with state-of-the-art methods for identifying peptidebinding motifs of aligned (exemplified by MHC class I) as well as unaligned (exemplified by MHC class II) quantitative peptide sequence data. Further, demonstrating the general utility of NNAlign, we have used it to characterize the cleavage specificities of proteases from high-throughput peptide array data. If a sufficient number of training examples can be generated, including negative instances, we could envision applying the method also on data generated by phage display peptide libraries. Other instances of recognition of short specific peptide motifs occurs frequently in biology where they are involved in molecular interaction, recognition, signaling, internalization, modification etc (e.g. phosphorylation, dephosphorylation, trafficking motifs, SH2 and SH3 domains, glycosylation, lipidation, etc. In contrast to domain recognition, short linear peptide sequences are thought to be particularly difficult to identify due to their unordered structure [40] . NNAlign appears to be ideally suited to identify such short linear peptide targets. Due to its simple interface and robust performance, we believe the method to constitute a significant tool providing the non-bioinformatician end-user with the ability to perform advanced bioinformatics-driven analysis of largescale peptide data sets. The data set of quantitative peptide-MHC class I binding affinity data published by Peters et al. [27] contains data from 48 different human, mouse, macaque and chimpanzee alleles. We selected 12 representative human alleles, and extracted binding data for 9-mer peptides maintaining the subsets of the original benchmark. This allows comparing the performance of NNAlign to the other methods presented in the paper by Peters et al. Figure 5 . Analysing high-density peptide array data with NNAlign. a) Fluorescence microscopy picture of a peptide microarray. The image is a magnified segment of the peptide chip used in the trypsin cleavage analysis. b) Trypsin peptide-chip data. The normalized observed (target) likelihood of cleavage as a function of the prediction score for the trypsin data set. Localizations of peptides containing the pairs of amino acids RP, RA or RR are highlighted in the plot. Proline (P) is known to prevent cleavage after arginine (R), whereas cleavage is observed with other amino acids such as R and A. c) Chymotrypsin peptide-chip data. Correlation plot between predicted and measured (target) data from the chymotrypsin data set. Values are binned by their x,y proximity, so that the scatterplot represents the density of data in each bin. NNAlign was trained with linear rescaling of the quantitative data, a motif length of 4 amino acids without inclusion of PFR encoding, Blosum encoding of peptide sequences, a combination of 3,7,15 hidden neurons, 10 initial seeds, 5-fold exhaustive cross-validation, training was stopped on the best test set performance. doi:10.1371/journal.pone.0026781.g005 MHC class II data set A large set of over 17,000 HLA-peptide binding affinities was published by Wang et al. [25] containing data from several different human alleles including HLA DR, DP and DQ alleles. For each allele, the predictive performance of various methods was estimated on the similarity reduced (SR) data set, where sequence similarity is minimized in order to avoid overlap between crossvalidation subsets. We preserved the same subsets for our crossvalidation, for easy comparison of the results and predictive performances. Peptide arrays were synthesized by Schafer-N, Copenhagen, Denmark using a maskless photolithographic technique [41] in which 365 nm light is projected onto NPPOC-photoprotected [42, 43] amino groups on a glass surface in patterns corresponding to the synthesis fields. Details of the technique will be published elsewhere, but briefly, the patterns were generated using digital micromirrors and projected onto the synthesis surface using UVimaging optics. In each layer of amino acids, the relevant amino acids were coupled successively to predefined fields after UVinduced removal of the photoprotection groups in those fields. The couplings were made using standard Fmoc-amino acids activated with HBTU/DIEA in NMP. After coupling of the last Fmocamino acid in each layer, all Fmoc-groups were removed in 20% piperidine in NMP and replaced by NPPOC groups coupled as the chloroformate in DCM with 0.1 M DIEA. The procedure was then repeated until all amino acids had been added to the growing peptide chains. Final cleavage of side protection groups was performed in TFA:1,2-ethanedithiol:water 94:2:4 v/v/v for 2 h at room temperature. Trypsin data. Peptide arrays were incubated for 30 min at room temperature with 0.1 g/L bovine Trypsin (Sigma T9201) dissolved in 0.1 M Tris/Acetate pH 8.0. After washing in the same buffer containing 0.1% SDS, the slides were washed with deionized water and air-dried. Staining of amino groups exposed by enzyme cleavage was made by incubation the slide for 30 min in 0.1 mg/mL Dylight549-NHS (Thermo Scientific) in 9:1 v/v nmethyl pyrrolidone:0.1M n-methyl morpholine/HCl pH 8 for 10 minutes. Chymotrypsin. Peptide arrays were incubated for 30 min at room temperature with 0.1 g/L bovine Chymotrypsin (Sigma C4129) dissolved in 0.1 M Tris/Acetate pH 8.0. After washing in the same buffer containing 0.1% SDS, the slides were washed with deionized water and air-dried. Staining of amino groups exposed by enzyme cleavage was made by incubation of the slides for 10 min in 1 mM 5(6)-TAMRA (carboxytetramethylrhodamine, Fluka 21953) activated with 1 eq HBTU, 2 eq DIEA in nmethylpyrrolidone. Recording of signals from peptide arrays. After incubation with activated fluorochromes, the peptide array slides were washed in the incubation buffer without fluorochrome followed by washings in n-methylpyrrolidone and dichloromethane and airdried. Images of the arrays were recorded using a MVX10 microscope equipped with a MT10_D fluorescence illumination system and a XM10 CCD camera (all from Olympus). The excitation wavelength was 530-550 nm and the emission filter was 575-625 nm. The images were analyzed using the PepArray analysis program (Schafer-N, Copenhagen Denmark). Data pre-processing. The quantitative peptide data entered by the user is rescaled to be between 0 and 1 before being fed to the neural network. The user is also given the option to apply a logarithmic transformation to the raw data, if its distribution appears to be too squashed towards low values. Outliers deviating more than 3 standard deviations from the average, which after rescaling would produce sparse regions in the spectrum with no data, are set at a value of exactly 3 standard deviations. This procedure produces ideal data for artificial neural network (ANN) training, with all values in the range [0:1] and the bulk of the data in the central region of the spectrum. The parameters for the rescaling function are defined separately on each of the training sets used in cross-validation, and then also applied to rescale their relative test sets. Subsets for cross-validation. In a n-fold cross-validation, n subsets are created from the complete dataset, and at each step n-1 subsets are used for training and 1 subset for testing. NNAlign offers three alternatives to create the subsets: i) random, splits the data into n subsets randomly; ii) homology clustering, uses a Hobohm 1 algorithm [44] to identify sequences that share an ungapped alignment with more than a specified fraction of matches; iii) common motif clustering, looks for stretches of identical amino acid between pairs of sequences as described by Nielsen et al. [19] . For both methods ii) and iii) similar sequences are grouped together in the same subset, but it is possible to choose to only include one representative for each group and disregard the other sequences from training. In this phase, if the input data contains repeated flanks (as might be the case in peptide array experiments, where linker sequences can be attached at the extremities of all peptides), these flanks are discarded, as they would affect the overlap estimation. If the user reckons that the repeated flanks might contain meaningful biological signal, an option allows retaining them in the training data. Note that in common motif clustering, the motif length is taken as the smallest in the interval of length given by the user. Thus, depending on the selected interval the subsets might be constructed in a different way and that could influence the cross-validated performance. Neural network training. The neural network training is performed as described by Nielsen et al. [20] . Initially, all network weights are assigned random values. From the current network configuration, the method selects the optimal n-mer core (and potential peptide flanking residues) for each of the peptides within the training set. The network weights are next updated, to lower the sum of squared errors between the observed and predicted score, the cores are redefined based on the new network configuration, and the procedure is iterated. An ensemble of ANNs is trained on the cross-validation subsets, with architecture parameters specified by the user. The motif length, encoding of flanks and peptide length determine the size of the input layer. If the motif length is given as an interval of values, multiple runs of ANN training are performed on the different lengths, and the length that produces the best cross-validated performance in terms of root mean square error (RMSE) is chosen for the final ensemble. The number of hidden neurons may be specified as a list of multiple values, so that an ensemble of networks is constructed with hidden layers of different sizes. Each architecture is trained multiple times, starting from different initial random configurations, to avoid as much as possible choosing suboptimal solutions producing local minima. Sequences can be presented to the network either with Sparse or Blosum encoding. In Sparse encoding, a vector of length N represents each amino acid, where all values are identical apart from the one representing the observed amino acid. Blosum encoding, on the other hand, takes into account amino acids similarity and partially allows substitutions of similar amino acids while penalizing very dissimilar ones [45] . Performance measures. Cross-validation allows estimating a method performance without the need of external evaluation data. The subsets reserved as test-sets are run through the network trained in the same cross-validation step, and Pearson's correlation, RMSE and Spearman correlation are calculated between observed and predicted values. It is possible to use the internal subsets to stop the training phase on the best test set performance in terms of RMSE. In this mode, performance can be estimated in an exhaustive or in a fast way. Exhaustive n-fold cross-validation (CV) consists of a nested CV procedure. At each step, 1 subset is left out as evaluation set, and the remaining subsets are used to generate a network ensemble in an n-1 CV training. In this CV training, the selected network configuration is the one that gives the minimum RMSE on the stopping set. Next the predictions for the evaluation data are estimated as a simple average of the prediction values for each network in the training ensemble. The exhaustive CV procedure adds one level to the cross-validation and increases greatly the running time. In alternative, the fast evaluation skips one nested level by using the same subset for stopping and evaluating performance, for a quicker but likely less accurate performance estimation. Final network ensemble. With cross-validated ANN training, each network has been evaluated on data not included in the training. The networks can then be ranked by performance, and only the top N for each cross-validation step will be included in the final ensemble, with N specified by the user. The final network ensemble can be downloaded to local disk, and used for predictions on new data by loading it to the NNAlign server submission page. Sequence motif logo. A list of 100,000 random naturally occurring peptides with length L = motif length+2 * flank length, generated from random UniProt [46] sequences, is presented to the individual networks in the ensemble. For each network, the 1% peptides that obtain the highest prediction scores are used to create a position specific scoring matrix (PSSM) that represents the motif captures by the neural network. Using a Gibbs sampler approach, all PSSMs are aligned to maximize the information content of the combined matrix. This ''offset correction'' step is obtained by repeatedly attempting to shift the starting position of randomly chosen PSSMs, and accepting/rejecting the move according to the conventional Metropolis Monte Carlo probability relation [47] : Where DI is the change in information content between the new and old offset configuration and T is a scalar that is lowered during the calculation. The process assigns to each PSSM, and to its relative network, an offset value that quantifies the shift distance from other networks. The re-aligned cores from the 1% scoring of 100,000 peptides are finally used to generate a combined sequence logo with the WebLogo program [48] . The offset correction can be skipped if the user chooses to, and in this case the logo is simply created by presenting the list of random peptides to the ANN final ensemble and selecting the 1% peptides that obtain the overall best score. Evaluation data. Additional data not included in the training can be uploaded to the NNAlign Server as an evaluation set. Evaluation data must be a list of peptides, with or without associated values, or a file in FASTA format. In the first case, all peptides are run through the trained network ensemble, and scored accordingly to their best alignment core. If values are provided together with peptides, they are assumed to be target values for validation purposes, and statistical measures between these values and predictions are calculated. In the case a FASTA file is loaded as evaluation set, the sequences therein contained are cut into peptides of length L = motif length+2 * flank length, shifting the starting position of one amino acid at a time. The generated peptides are all fed to the network to identify those that most closely match the motif learned by the ANNs. The results are sorted by prediction value, so that the best candidates are displayed at the top of the list. Sequence logos were introduced by Schneider et al. [49] as a way to represent graphically the pattern in a set of aligned sequences. The height R i of each column i in the logo is given as the information content in bits of the alignment at that particular position, and for a sufficiently large number of sequences and a 20letter alphabet it is calculated as:

Red blood cell transfusion in the critically ill patient

Red blood cell (RBC) transfusion is a common intervention in intensive care unit (ICU) patients. Anemia is frequent in this population and is associated with poor outcomes, especially in patients with ischemic heart disease. Although blood transfusions are generally given to improve tissue oxygenation, they do not systematically increase oxygen consumption and effects on oxygen delivery are not always very impressive. Blood transfusion may be lifesaving in some circumstances, but many studies have reported increased morbidity and mortality in transfused patients. This review focuses on some important aspects of RBC transfusion in the ICU, including physiologic considerations, a brief description of serious infectious and noninfectious hazards of transfusion, and the effects of RBC storage lesions. Emphasis is placed on the importance of personalizing blood transfusion according to physiological endpoints rather than arbitrary thresholds.

Red blood cell (RBC) transfusion is commonly required in critically ill patients. Several recent, observational, multicenter studies reported that approximately one third of critically ill patients received a blood transfusion at one time or another during their stay in the intensive care unit (ICU) ( Table 1) . Because of the frequent use of this intervention, it is important for the ICU physician to be aware of recent developments in this continuously evolving field of medicine. In this narrative review, we consider some key aspects of transfusion medicine in the ICU, focusing on aspects relevant to the critically ill patient, including prevalence and reasons for blood transfusion, epidemiology and etiology of anemia in these patients, pathophysiological considerations on tolerance to anemia, and efficacy of RBC transfusion. Safety concerns, including questions of RBC storage and leukoreduction, are then discussed, followed by a proposal for an integrated approach to transfusion decisions and a discussion on economic aspects and alternatives to blood transfusion. Anemia is common in ICU patients and appears early in the ICU course [1] . In an observational, multicenter, cohort study in Scotland, 25% of patients admitted to the ICU had a hemoglobin level < 9 g/dl [2] . Similar results were reported in the ABC study [3] , in which 29% of patients had a hemoglobin concentration < 10 g/ dl on admission. Even in nonbleeding ICU patients, hemoglobin levels tend to decrease early [3] . This decrease is more pronounced in septic than in nonseptic patients [4] , at least in part because of their inflammatory response; more frequent blood sampling may also contribute. Interestingly, anemia and the need to restore adequate oxygen delivery (DO 2 ) are the most common indications for transfusion, rather than acute bleeding [3, [5] [6] [7] [8] [9] [10] . Anemia in the critically ill patient is a multifactorial phenomenon that has been compared to the so-called "anemia of chronic illness" [11] . Apart from evident causes of anemia, such as primary blood losses (e.g., trauma, surgery, gastrointestinal bleeding), multiple other etiologies contribute to its pathophysiology and often coexist in the same patient [11] . These include blood losses related to minor procedures or phlebotomy, and hemodilution secondary to fluid resuscitation. Some studies have suggested that blood sampling may average as much as 40 ml/day [3, 4] , but the amount of blood required may decrease with technological developments in analytic methods. Other mechanisms for anemia include an inflammatory response with blunted erythropoietin (EPO) production, abnormalities in iron metabolism, and altered proliferation and differentiation of medullar erythroid precursors [11] . As a consequence, RBC deformability is decreased [12] , whereas RBC adherence to the endothelium is increased, especially in septic patients, potentially leading to microcirculatory impairment and tissue hypoxia [13] . Tolerance to anemia in healthy subjects and in the critically ill patient Tolerance to anemia is highly dependent on the volume status of the patient, physiological reserve, and the dynamics of the anemia (for example, chronic, such as the anemia of sepsis, versus acute, such as hemorrhagic conditions). Normovolemic anemia is better tolerated than anemia in hypovolemic states (e.g., acute bleeding in trauma patients or surgery) in which cardiac output acutely decreases. In healthy subjects submitted to normovolemic hemodilution, cardiac output increases because of decreased blood viscosity (especially relevant in severe anemia) and increased adrenergic response, allowing tachycardia and increased myocardial contractility. Other phenomena include blood flow redistribution (to heart and brain) and an increased oxygen extraction ratio (reflected by a decrease in mixed venous saturation [SvO 2 ]). These mechanisms allow healthy humans to tolerate severe degrees of normovolemic anemia [14, 15] , although side effects, such as arrhythmias or ST changes, can be observed in extreme cases [16, 17] . The myocardium is the organ at risk in cases of acute anemia in which both tachycardia and increased ventricle contractility may increase myocardial oxygen demand. Because myocardial oxygen extraction is already almost maximal at rest, every increase in myocardial oxygen demand must be accompanied by increased coronary blood flow [18] . This can become problematic in patients with stenotic coronary arteries especially when tachycardia is present, which can decrease diastoledependent left ventricle perfusion. Therefore, in critically ill patients, especially those with heart failure or coronary artery disease (CAD), the myocardium may not tolerate such low hemoglobin levels [19] . In acute myocardial infarction, anemia may worsen myocardial ischemia, generate arrhythmias, and potentially increase infarct size [20] . In patients with acute coronary syndrome or heart failure, anemia increases morbidity and mortality [21, 22] . For these reasons, patients with cardiac problems should be managed with a more liberal approach to transfusion than other patients [23, 24] . The primary purpose of blood transfusion is to increase DO 2 , which is determined by cardiac output and arterial content of oxygen, the latter being dependent on the hemoglobin level. Hence, blood transfusions can, theoretically at least, limit tissue hypoxia [13, 25, 26] . But does this really happen in clinical practice? It is obvious that RBC transfusions can be lifesaving in situations of acute severe anemia or in bleeding patients in whom RBC administration can increase both oxygen arterial content and cardiac output. However, in the absence of bleeding, the increase in hemoglobin concentration could very well be offset by a decrease in cardiac output because of the increase in blood viscosity associated with a decreased sympathetic response [27, 28] . DO 2 has been shown to increase following RBC transfusion in numerous studies [26] , but not in all [29] . The effects of RBC transfusion on the relationship between DO 2 and oxygen uptake (VO 2 ) are even more difficult to predict. Some studies reported that VO 2 increased following RBC transfusion, whereas others did not [26] , and variable effects have been reported on tissue perfusion as assessed by gastric mucosal pH or near-infrared spectroscopy (NIRS) [30] . The reasons for these contradictory results lie primarily in the degree of severity of hypoxia preceding the RBC transfusion [31] , which influences the dependency of VO 2 on DO 2 . Methodological problems (imprecision in determination of VO 2 , assessment of global VO 2 instead of regional VO 2 , poor correlation between systemic oxygenation parameters, and oxygenation in the microcirculation [13] ) also may contribute to these discrepancies [31] . Red blood cell transfusions have been associated with worse outcomes in several populations of patients, including critically ill patients. In a recent systematic review of 45 observational studies reporting the impact of transfusions on patient outcome (mortality, infections, acute respiratory distress syndrome [ARDS]) in populations of trauma, general surgery, orthopedic surgery, acute coronary syndrome, and ICU patients, Marik and Corwin [32] identified RBC transfusion as an independent predictor of death (pooled odds ratio (OR) from 12 studies, 1.7; 95% confidence interval (CI), 1.4-1.9), infectious complications (pooled OR from 9 studies, 1.8; 95% CI, 1.5-2.2), and ARDS (pooled OR from 6 studies, 2.5; 95% CI, 1.6-3.3). In ICU patients, the three studies included in the review (ABC study [3] , CRIT study [5] , and a study by Gong et al. [33] ) consistently showed a statistically significant association of RBC transfusion with mortality. On the other hand, analysis of data from a multicenter, prospective, observational study of 3,147 patients in 198 European ICUs (the SOAP study) indicated that blood transfusions were not associated with increased mortality by multivariate analysis or propensity matching [34] . In contrast, an extended Cox proportional hazard analysis showed that patients who received a transfusion in fact had a better survival, all factors being otherwise equal. An increased rate of transfused leukoreduced RBCs reported in this study (in which 76% of centers routinely used leukoreduced RBCs) could perhaps account for the differences between the earlier ABC study [3] (in which 46% of centers used leukodepleted blood most of the time) and the SOAP study [34] . It also is possible that transfusion thresholds have become so low that the benefits of blood transfusion outweigh the risks. In patients with acute coronary syndrome, several studies have shown poorer outcomes, including increased mortality, in transfused groups compared with nontransfused patients after adjustment for potential confounders [21, [35] [36] [37] ; similar findings have been reported in patients who undergo percutaneous coronary interventions (PCI) [38] . However, although still controversial, RBC transfusions may be useful in subgroups of elderly patients with acute myocardial infarction [39] or patients with ST elevation myocardial infarction (STEMI) [21] . Patients who undergo cardiac surgery seem to have worse outcomes when transfused, including higher mortality [40, 41] , increased occurrence of postoperative infections [41, 42] , increased time on mechanical ventilation [40, 43] , and higher incidence of postoperative acute kidney injury [41, 44] . Other studies have reported that trauma patients [45, 46] , including those with burns [47] , may have increased mortality rates associated with receiving blood transfusions. In contrast, RBC transfusion has been reported to be associated with improved outcomes in patients with traumatic brain injury or subarachnoid hemorrhage [48, 49] . In the early resuscitation of patients with severe sepsis, implementation of a therapeutic protocol that included RBC transfusion to obtain a hematocrit > 30% was associated with a significant reduction in hospital mortality [50] . These results should be interpreted with caution, because most of these data come from observational, retrospective studies, which are subject to numerous biases and sometimes control poorly for confounders, despite the use of various statistical tools, such as logistic regression [51] . It is clear that analyses should not include only admission data. For example, in a welldefined patient population, such as after cardiac surgery, patients who develop gastrointestinal bleeding and require a blood transfusion have a worse prognosis, which is not necessarily the result of the blood transfusion. It is of paramount importance that all risks factors are taken into account. Ruttinger et al. [52] illustrated this point very well. In a series of more than 3,000 surgical patients, these authors showed by using a limited multivariable analysis that transfusions were associated with a worse outcome, but a more complete analysis cancelled out this statistical observation. The reasons for the apparent worse outcome of transfused compared with nontransfused critically ill patients may be found in several detrimental effects of transfused blood, globally referred to under the acronym "Non-Infectious Serious Hazards Of Transfusion" or NISHOT ( Table 2 ) [53] . These include, among others, deleterious effects on the immune system (transfusion-related immunomodulation or "TRIM") or on the cardiopulmonary system, e.g., transfusion-related acute lung injury ("TRALI") [54] or transfusion-associated circulatory overload ("TACO"); the latter is currently the leading reported cause of transfusion-associated mortality [55] . These effects may be enhanced by pathologic conditions (e.g., sepsis) in which the microcirculation is impaired [56] and/or when the RBCs have been stored for some time. During storage, RBCs undergo a series of biological and biochemical changes collectively referred to as "the storage lesion" [57] . This includes intracellular changes (progressive depletion of 2,3-diphosphoglycerate [2,3-DPG] with increased affinity of hemoglobin for oxygen, depletion of ATP), membrane changes (membrane vesiculation, morphological changes eventually leading to irreversibly deformed spheroechinocytes, lipid peroxidation and increased expression of phosphatidylserine, decreased deformability), and changes in the storage medium (decreased pH, increased potassium, release of proinflammatory cytokines). These stored RBCs also have an increased tendency to adhere to endothelium and could promote vasoconstriction; the stored RBCs act as a "sink" for nitric oxide [58] . Some animal studies [13] have shown deleterious effects of old RBCs on the microcirculation (potentially leading to tissue hypoxia and organ dysfunction). A human study found an inverse correlation between the age of transfused RBCs and maximal change in gastric mucosal pH, but these findings were challenged in subsequent studies [59] [60] [61] . The clinical consequences of storage lesions are still not clear. A recent review of the literature [57] identified 24 studies that address the effects of RBC length of storage on clinical (mortality, infections, length of stay, length of mechanical ventilation) or physiological (microcirculation, gastric mucosal pH) endpoints. Some studies found associations between the age of transfused RBCs and poorer outcomes, whereas others did not. Overall, no clear detrimental effect of RBC age could be identified; however, definitive conclusions are difficult to obtain because of numerous statistical limitations and biases inherent to the study designs [51, 62] . Several, large, randomized, controlled trials in adult ICU and cardiac surgery patients are currently ongoing to address the clinical relevance of RBC storage. In the multicenter, double-blind prospective ABLE (Age of Blood Evaluation) study [63] , adult patients admitted to the ICU are randomly assigned to receive leukoreduced RBCs stored for less than 7 days or issued according to standard procedure (expected average storage time of 19 days). The primary endpoint of this study is 90-day all-cause mortality. The target number of patients is 2,510 (for an expected improvement in primary endpoint greater than 5%) with an anticipated completion date by April 2013. The Red Cell Storage Duration Study (RECESS) is a multicenter, randomized study in patients (age 12 years and older) who undergo complex cardiac surgery and are likely to require RBC transfusion [64] . Patients who need transfusion are randomized to receive RBCs stored for ≤ 10 days or ≥ 21 days. The primary endpoint of this study is the change in the Multiple Organ Dysfunction Score (MODS) from baseline to day 7, with secondary outcomes including all-cause 28-day mortality. The target number of patients is 1,832, and the anticipated completion date is September 2013. The results of these trials, especially if older blood appears to be harmful, could have important logistic implications for blood banks [65, 66] . Many of the adverse effects associated with the transfusion of allogeneic RBCs have been shown to be related to the infusion of white blood cells (WBCs) present in the blood product. Leukoreduction is a process in which WBCs are reduced in number through centrifugation or filtration [67] . This process allows removal of approximately 99.995% of WBCs, but several thousand leukocytes (0.005% of a 500 ml blood unit) may still be present in the processed blood [67] ; hence, the word "leukoreduction" is better than "deleukocytation." The beneficial effects of this process include decreased [67] , and possibly decreased lung injury, such as TRALI. Moreover, prestorage leukoreduction, in which WBC removal occurs before RBC storage, avoids the need for a leukodepletion filter during transfusion [67] (but a 170-200-μm filter still needs to be incorporated into the intravenous blood line). In several studies, prestorage leukoreduction decreased RBC storage lesions, with fewer immunomodulating properties [68] and less adhesion of stored RBCs to the endothelium [69] . A clinical benefit of leukoreduction is still somewhat controversial, particularly in the critically ill patient where no randomized, controlled trial has been performed [70] . In a before-after study of 14,786 patients who underwent cardiac surgery, repair of hip fracture, or who required intensive care after surgery, there was a 1% decrease in mortality rate associated with the implementation of universal leukoreduction [71] . In a recent meta-analysis of nine RCTs involving 3,093 surgical patients, the use of leukoreduction significantly reduced the odds of postoperative infection (summary OR, 0.522; 95% CI, 0.332-0.821; p = 0.005) [72] . This observation had been suggested in a previous meta-analysis [73] but has been challenged by another recent meta-analysis [74] . Nevertheless, leukoreduction makes sense, and many countries have adopted it as routine, even though costs are elevated. In Europe, at the time of the SOAP study in 2002, 76% of centers reported using leukodepleted blood routinely [34] , whereas an earlier study performed in the same countries reported lower rates [3] . Classically, the decision to transfuse is driven by arbitrary "triggers" (hemoglobin level) rather than clinical or physiologic findings. Data from the CRIT study [5] , in which there was little evidence that age or comorbidities significantly influenced transfusion practice, tend to support this view. Current recommendations for RBC transfusion [75, 76] are mainly based on the famous "TRICC" (Transfusion Requirements In Critical Care) trial in which patients assigned to a restrictive transfusion strategy (transfusion if hemoglobin level < 7 g/dl) had similar 30-day mortality rates (and even lower mortality in subgroups with APACHE II < 20 and patients younger than age 55 years) than patients transfused according to a more liberal strategy (transfusion if hemoglobin level < 10 g/dl) [77] . In cardiac surgery patients, the recent randomized, monocenter "TRACS" (Transfusion Requirements after Cardiac Surgery) trial, which compared a restrictive to a liberal strategy (transfusion when hematocrit < 24% or < 30%, respectively), reported no difference in the primary endpoint (composite of 30-day mortality and morbidity [cardiogenic shock, ARDS, acute kidney injury]) between the groups [78] . However, it is quite clear there is no "magic" hemoglobin or hematocrit trigger, and for the same level of hemoglobin, some patients will do well, whereas others will not. Thus, the decision to transfuse a patient should be individualized, taking into account several factors, including signs and symptoms of tissue hypoxia (angina pectoris, cognitive dysfunction diagnosed by neuropsychological tests, or increased P300 latencies [79] [80] [81] ), increased blood lactate levels [82] , or electrocardiographic changes suggestive of myocardial ischemia. Indirect measures of oxygenation, such as a decreased SvO 2 or central venous oxygen saturation (ScvO 2 ), also may be considered [82] . For example, in a study of early goal-directed therapy in patients with severe sepsis or septic shock admitted to an emergency department, a decrease in ScvO 2 < 70% initiated a therapeutic intervention, including fluid resuscitation, inotropes, vasopressors, and RBC transfusion to increase hematocrit to > 30% [50] . Use of a decreased ratio of cardiac index to oxygen extraction (CI/EO 2 ratio) may be better, because this parameter also reflects the cardiac response to anemia [83] . The costs of blood transfusion are particularly complex to assess because of the many factors that have to be taken into consideration (blood collection and screening for pathogens; blood component processing, including leukoreduction, storage, transport to the transfusion facility; administration of blood to the patient; management of potential short-and long-term transfusion-related side effects) [84] . The subtype of the blood unit also may play a role because some products, such as CMV-negative or autologous units, are costlier than classical allogeneic RBCs. Consequently, studies in this field have given extremely varied results, which are not easily comparable. Evidence has shown increased costs of RBC transfusion over time [85] , related to various factors, including (but not limited to) use of leukoreduction and more sophisticated methods for pathogen detection, such as nucleic acid testing (NAT) [84] . For example, a study in Canada evaluated the mean societal cost of one allogeneic RBC unit at 264.81 US$, twice the cost estimated 7 years earlier [86] . Generally, these reported values are probably underestimated, and some have calculated that the cost of blood to society could in fact be twofold higher [84] . Because of limited availability, costs and safety concerns related to blood transfusion, several strategies to reduce blood transfusions can be considered in addition to increasing transfusion trigger thresholds. These include approaches to reduce blood losses, for example use of antifibrinolytic agents, such as tranexamic acid or epsilon-aminocaproic acid (EACA) and techniques of cell salvage during surgery; also, the use of small volume sample tubes can limit the blood losses related to sampling for laboratory studies. In a meta-analysis of 9 randomized controlled trials [87] , subcutaneous administration of recombinant erythropoietin (EPO) in critically ill patients was shown to be associated with decreased transfusion rates, but this was not associated with improved mortality (except possibly in a subgroup of trauma patients [88] ). Concerns also have been raised about potentially increased rates of deep vein thrombosis [88] . The development of artificial oxygen carriers is under investigation, but these have their own problems [89] . Further research is needed to improve these alternative strategies. RBC transfusion can be lifesaving. During the past two decades, however, safety concerns have emerged, with suggestions that morbidity and mortality may be increased in patients who receive blood transfusions. Therefore, the decision to transfuse should be individualized, based on a rational approach and taking into account physiologic variables in addition to the hemoglobin value. This strategy, along with the use of alternatives whenever possible to limit bleeding, should limit unnecessary exposure to RBCs. Authors' contributions CL drafted the manuscript. The manuscript was revised for intellectual content by JLV. Both authors read and approved the final manuscript.

Respiratory support by neurally adjusted ventilatory assist (NAVA) in severe RSV-related bronchiolitis: a case series report

BACKGROUND: Neurally adjusted ventilatory assist (NAVA) is a new mode of mechanical ventilation controlled by diaphragmatic electrical signals. The electrical signals allow synchronization of ventilation to spontaneous breathing efforts of a child, as well as permitting pressure assistance proportional to the electrical signal. NAVA provides equally fine synchronization of respiratory support and pressure assistance varying with the needs of the child. NAVA has mainly been studied in children who underwent cardiac surgery during the period of weaning from a respirator. CASE PRESENTATION: We report here a series of 3 children (1 month, 3 years, and 28 days old) with severe respiratory distress due to RSV-related bronchiolitis requiring invasive mechanical ventilation with a high level of oxygen (FiO(2 )≥50%) for whom NAVA facilitated respiratory support. One of these children had diagnosis criteria for acute lung injury, another for acute respiratory distress syndrome. Establishment of NAVA provided synchronization of mechanical ventilatory support with the breathing efforts of the children. Respiratory rate and inspiratory pressure became extremely variable, varying at each cycle, while children were breathing easily and smoothly. All three children demonstrated less oxygen requirements after introducing NAVA (57 ± 6% to 42 ± 18%). This improvement was observed while peak airway pressure decreased (28 ± 3 to 15 ± 5 cm H(2)O). In one child, NAVA facilitated the management of acute respiratory distress syndrome with extensive subcutaneous emphysema. CONCLUSIONS: Our findings highlight the feasibility and benefit of NAVA in children with severe RSV-related bronchiolitis. NAVA provides a less aggressive ventilation requiring lower inspiratory pressures with good results for oxygenation and more comfort for the children.

Neurally adjusted ventilatory assist (NAVA) is a new method of assisted ventilation that can be used for children regardless of weight and age. This ventilation mode is controlled by diaphragmatic electrical signals through a gastric tube with specific electrodes on its surface. The collected electrical signals allow synchronization of ventilation to spontaneous breathing efforts of a child, as well as providing pressure assistance proportional to the electrical signal and thus to the output of the child's respiratory centers. NAVA has mainly been studied in children who underwent cardiac surgery [1] [2] [3] during the period of weaning from a respirator. We report here a series of 3 children with severe respiratory distress due to respiratory syncytial virus (RSV) bronchiolitis for whom NAVA facilitated respiratory support. Our unit is a 12-bed tertiary care university hospital pediatric intensive care unit. Recruitment is both medical and surgical. NAVA has been used in our unit for weaning children from a respirator who were operated * Correspondence: jeanmichel.liet@chu-nantes.fr † Contributed equally 1 Unité de Réanimation Pédiatrique, Hôpital Mère-Enfant Faïencerie, CHU de Nantes, 38 Boulevard Jean-Monnet, 44093 Nantes, France Full list of author information is available at the end of the article on for congenital heart disease. The effectiveness of NAVA in these children led us to gradually expand the indications. We report a series of 3 children with severe respiratory distress due to RSV bronchiolitis for whom NAVA was used. The local ethics committee (groupe nantais d'éthique dans le domaine de la santé [GNEDS]) considered our report as non-interventional data research. The parents of all three children gave their written consent for publication. Starting NAVA requires the initial correct positioning of the "NAVA" gastric tube. This commercially available feeding tube equipped with sensors (Edi catheter, Maquet Critical Care, Solna, Sweden) permits the recording of electrical activity of the diaphragm (Edi) via a Servo-I Ventilator (Maquet Critical Care, Solna, Sweden) using a standardized method [4] . Settings are relatively simple and include positive end-expiratory pressure (PEEP), fraction of inspired oxygen (FiO 2 ), and level of NAVA assistance. The Edi was multiplied was multiplied by the NAVA level to adjust the pressure assistance delivered to the child. The delivered pressure is equal to: NAVA level × (Edi max -Edi min) + PEEP. In clinical practice we usually started with a NAVA level of 1 cm H 2 O/μV that may have required adjustment if the Edi max signals deviated from a range between 5 and 20 μV. If the Edi signals turned out to be consistently greater than 20 μV, we increased the NAVA level until they are within this range. In the three reported cases, we did not need to do so. During NAVA, the ventilator is triggered when the deflection in the Edi curve exceeds 0.5 μV. The assist is cycled-off when the Edi decreases to 70% of its peak value. We assume that pressure support, which is pneumatically triggered, should remain a means of backup ventilation in case the Edi signal cannot be collected (e.g. if the child removes his/her Edi catheter). Therefore, we set the trigger of this pressure support high enough (typically 0 to -5 cm) so that this backup ventilation did not compete with NAVA ventilation. We measured respiratory parameters (FiO 2 , tidal volume, mean airway pressure, peak inspiratory pressure, respiratory rate, and Edi max) directly from data exported from the respirator. Nurses recorded vital signs and SpO2. The oxygenation saturation index, OSI = (FiO 2 × mean airway pressure)/SpO 2 , was used to provide a non-invasive method of oxygenation assessment. This index can be used for the diagnosis of acute lung injury (ALI) and acute respiratory distress syndrome (ARDS) in children when SpO 2 values are ≤ 97% [5] . Diagnosis of ALI and ARDS required an acute onset of the process, bilateral infiltrates on a chest radiograph, no evidence of left atrial hypertension, and OSI > 6.5 (ALI) or > 7.8 (ARDS). Blood samples were also analyzed to provide blood pH and PCO 2 values. Shemsy was one month old (3.8 kg) without any particular risk factors. Her parents referred her to the emergency room because of grunting and hypotonia. She had rhinitis for several days with a cough for 48 hours following a viral contamination 3 days prior. She presented with apnea, desaturation, bradycardia and altered consciousness. She was intubated and ventilated immediately, and then transferred to intensive care. Ten hours later, respiratory parameters were as follows: synchronized intermittent mandatory ventilation (SIMV) with a tidal volume (VT) of 20 ml (5 ml/kg); rate, 30/min; PEEP, 5 cm H 2 O; and FiO 2 , 50%. Measured parameters (stable for 2 hours) included a SpO 2 of 91%, a mean airway pressure of 10 cm H 2 O, and a peak inspiratory pressure of 30 cm H 2 O (other parameters are also shown in Table 1 ). The OSI was 5.5. A chest X-ray showed poorly ventilated lungs with diffuse infiltrates. Since the child was agitated, the options for care were to increase sedation, or to attempt ventilation using NAVA. A brief test was undertaken to validate the use of NAVA, which proved successful. We chose to commence NAVA after a short period of decreased sedation (morphine was decreased to 8 μg/kg/h). Initial NAVA settings were PEEP, 5 cm H 2 O; NAVA level, 1 cm H 2 O/μV; and FiO 2 , initially 50% was then decreased by nurses to SpO 2 > 90%. We observed a dramatic decrease in inspiratory pressure with a reduced requirement for oxygen (Figures 1 and 2 ). Within several minutes, the child's breathing became much more harmonious and smoother, while her respiratory parameters showed large variations from one cycle to another. SpO 2 was96%, mean airway pressure was 6 cm H 2 O, peak inspiratory pressure was 10 cm H 2 O, and minute volume was 0.6 l/min. Twelve hours later, FiO 2 was decreased to 21% with a mean airway pressure of 6 cm H 2 O. Detailed ventilatory parameters are reported in Table 1 . Since respiratory parameters were very low (Peak inspiratory pressure < 12 cm H 2 O with FiO2 < 25%) and blood gas values was normal, we extubated the child (10:30). She needed nasal continuous positive airway pressure for 3 days after which she was able to leave intensive care. Tracheal aspirate was positive for RSV and Streptococcus pneumoniae. Matteo was 3 years old (14 kg), he was prematurely born (birth weight 1650 g) and he was mechanically ventilated in the neonatal period during 10 days for pulmonary hemorrhage. He was recently hospitalized because of subcutaneous emphysema, with signs of acute respiratory failure from RSV infection. Because of an increased need for oxygen, he was intubated and ventilated with an FiO 2 of 100%, and PEEP was 3. A chest X-ray did not show pneumothorax that could be drained. Tracheal aspirate was positive for RSV and Hemophilus influenzae. A few days later, because of emergence of an alveolar syndrome associated with persistence of subcutaneous emphysema, he was transferred to our unit for possible extracorporeal membrane oxygenation (ECMO). Upon arrival, he had an oxygenation saturation index of 9.4 as well as with the other criteria for ARDS (bilateral infiltrates on a chest radiograph and no evidence of left atrial hypertension). Respiratory parameters were an SIMV with a VT of 85 ml (6 ml/kg), the rate was 25/ min, PEEP was 6 cm H 2 O, and FiO 2 was 70%. SpO2 was 89%, mean airway pressure was 12 cm H 2 O, peak inspiratory pressure was 28 cm H 2 O, and minute volume was 2 l/min. Venous blood gas showed a pH of 7.32, and a PvCO 2 of 53 Torr (7.1 kPa). Because of a slight improvement, the child was not treated by ECMO. He underwent fibroscopy to eliminate the diagnosis of foreign body inhalation, which would have explained the subcutaneous emphysema. Forty-eight hours later, because of an increase in cough, cutaneous emphysema worsened and became diffuse (cervico-thoraco-abdominal) despite the reduction in PEEP to 4 cm H 2 O (FiO 2 60%). We decided to use NAVA to improve the synchronization of mechanical ventilation with the child's spontaneous breathing. Sedation was reduced, midazolam was decreased to 80 μg/kg/h and sufentanil to 0.3 μg/kg/h. The child became alert. Initial NAVA settings were a PEEP of 5 cm H 2 O, NAVA level was 1.2 cm H 2 O/μV, and FiO 2 was 60%. This change reduced the requirement for oxygen and normalized blood gases ( Table 2 ). The next day, the subcutaneous emphysema started to decline ( Figure 3 ). Peak inspiratory pressure was between 10 and 19 cm H 2 O with an FiO 2 of 50% for a SpO 2 of 90%. The child's level of distress was scored with the modified COMFORT scale [6] and it ranged between 11 and 14 (adequately sedated, as confirmed by the child). This clear improvement allowed us to reassign the child to the original hospital 48 hours after initiation of NAVA. NAVA was continued at the second facility, permitting extubation 3 days later. Leane was 28 days old (3 kg) and was born at term. She had a 2-year-old brother with bronchiolitis. After an episode of rhinorrhea, she presented with feeding difficulties and was referred to the emergency room of a local hospital. Four hours later, she progressively presented with low oxygen saturation, tachypnea, and chest retraction. She was placed under nasal CPAP with 30% FiO 2 , and then transferred to our unit for severe RSV bronchiolitis. She was intubated on arrival because of clinical signs of respiratory distress and collapse. Although we suspected concomitant bacterial pneumonia because of a major inflammatory syndrome, we did not have bacteriological confirmation. Her respiratory status deteriorated rapidly. The OSI was 6.8 (with other diagnosis criteria for acute lung injury). Her need for oxygen increased rapidly with bilateral infiltrates on a chest radiograph. We chose to use NAVA after a short period of decreased sedation (morphine was decreased to 8 μg/kg/h). Initial NAVA settings were a PEEP of 5 cm H 2 O, NAVA level was 1 cm H 2 O/μV, and FiO 2 was initially 60% and was then adjusted by a nurse for a SpO 2 > 90% and < 98%. Six hours later, FiO 2 was 35%, while ventilatory pressures were lower than before starting NAVA ( Table 3) . As in the 2 other cases, ventilatory parameters were highly variable (Figure 4) , while chest movements of the child were smooth as if the child was not mechanically ventilated. Twenty-four hours after starting NAVA, FiO 2 was 21%. The modified COMFORT scale ranged between 7 and 13. At 36 hours of NAVA ventilation, the child was accidentally extubated during coughing. She immediately presented marked signs of respiratory distress (Silverman score: 7/10). We then used noninvasive ventilation with the NAVA option. Edi max values were initially very high (80 μV) and gradually decreased over 1 hour after the establishment of noninvasive ventilation. Thereafter, nasal continuous airway pressure was applied and the child left the intensive care unit 3 days later. As in many pediatric intensive care units, our rate of intubation of children hospitalized for bronchiolitis is tidal volume became much more variable from one cycle to another. Middle panels: One of the most remarkable changes observed with switching to the NAVA mode was the immediate reduction in the mean airway pressure (C) and in the peak airway pressure (D) which decreased from 30 to 10 cm H 2 O. Bottom panels: After starting NAVA, the respiratory rate became very variable over time (E). From a mandatory frequency set at 30 breaths per minute, respiratory rate increased to 40 and 60 breaths per minute. Clinically, the breathing became easier with harmonious chest movements. (F) Edi max that it is the sum of inspiratory Edi and Edi min corresponds to the peak of electrical activity of the diaphragm. In SIMV, this activity is depressed, and in NAVA, the inspiratory Edi (Edi max -Edi min) drives ventilation. Abbreviations: FiO 2 , fraction of inspired oxygen; Edi, electrical activity of the diaphragm; SIMV, synchronized intermittent mandatory ventilation. [5.7] Data in bold are prescribed settings. Other data are measured parameters that depend both on the ventilatory settings and the respiratory status of the child. * Data expressed in parentheses represent measurements that were very variable over time, and hence an estimate of the measured parameter is provided. Abbreviations: SIMV, synchronized intermittent mandatory ventilation; NAVA, neurally adjusted ventilatory assist; PEEP, positive end-expiratory pressure; FiO 2 , fraction of inspired oxygen; Edi, electrical activity of the diaphragm Figure 2 Screenshot with trends over 24 hours (case 1). In the window untitled "Courbes de tendances" (Trend curves), three panels report trends over 24 hours of peak inspiratory pressure (cm H 2 O), respiratory rate (resp/mn), and minute volume (l/min). On the right of these panels, the values of these ventilatory parameters were collected to the vertical bar (at 18:57 while the child was not receiving NAVA). The downward vertical arrow indicates the switch from SIMV to NAVA (at 20:30). Outside the window untitled "Courbes de tendances", on the right of the scrennshot, ventilatory parameters were collected the next day at 10:15 while the child was receiving NAVA. The upper pannel showed a decrease in peak inspiratory pressure after the switch of ventilation. The middle pannel showed the extreme variability of the respiratory ratein NAVA (the white area under the curve corresponds to the mandatory respiratory rate, while the black area corresponds to the spontaneous respiratory rate). The lower pannel showed minute volume that remained unchanged. When comparing values of ventilatory parameters in SIMV (at 18:57) with those in NAVA (next day at 10:15), peak inspiratory pressure decreased from 29 to 6 cm H 2 O, mean airway pressure decreased from 10 to 4 cm H 2 O, spontaneous respiratory rate varied from 0 to 29 breaths/min. Abbreviations: SIMV, synchronized intermittent mandatory ventilation; PEP, positive end-expiratory pressure; P crête, peak inspiratory pressure; P moyen, mean airway pressure; FR spont, spontaneous respiratory rate; F resp, respiratory rate; VM, minute volume; FiO 2 , fraction of inspired oxygen. low (< 20%) [7] . However, the severity of lung disease in some children still necessitates invasive mechanical ventilation. Our unit recruitment is both medical and surgical, and we therefore acquired expertise in NAVA through the weaning of children treated after cardiac surgery. We then expanded the indications of this mode of ventilation in children with severe respiratory disease. The positive results in the present study may be explained by the specific selection of children to whom we applied NAVA. First, there should be no specific contraindication to the placement of a nasogastric tube. Second, there should be no alkalosis or hypocapnia (in such cases there would not be sufficient diaphragmatic electrical activity). During alkalosis, the ventilatory brain centers no longer stimulate the diaphragm, and the respirator works on a backup mode, which is simply conventional ventilation. Finally, the level of sedation should not be too high, so that it does not depress brain centers that control breathing. If the sedation is too high, the Edi signal cannot be collected and the respirator works again in a backup mode. Likewise, neuromuscular connection from the respiratory center to the diaphragm must be intact. For example, NAVA cannot be used in case of post-surgical lesion of the two phrenic nerves [3] or diaphragmatic paralysis secondary to botulism [8] . The main benefit observed in our cases was an improvement in oxygenation associated with a normalization of blood pH. This was achieved with marked decrease in peak airway pressure. This effect has been previously found in crossover studies reporting NAVA in weaning children from a respirator who were operated on for congenital heart disease and comparing NAVA with pressure support [1] [2] [3] . Clinically, breathing becomes easier with harmonious chest movements. In one of our cases, NAVA was very effective for ventilation in a child who had both ARDS and extensive cutaneous emphysema. The excellent synchronization of mechanical ventilation with the spontaneous breathing of the child improved oxygenation without aggravating the emphysema. Several factors could explain the beneficial effects observed with NAVA in these three children who had severe respiratory distress. First, asynchrony is associated with increased morbidity, a longer duration of ventilation, and a longer hospital stay [9] [10] [11] . There are few pediatric data published regarding the adverse effects of long-term asynchrony between mandatory ventilation and the respiratory efforts of children, but it has been shown that infant-ventilator asynchrony (both inspiratory and expiratory asynchrony) may affect more than 50% of the total breath duration [12] . Second, NAVA provides assistance in synchronization, as well as in pressure assistance in Recording began with the establishment of NAVA. Upper panels: After starting NAVA, (A) FiO 2 gradually decreased to 35%, and (B) tidal volume became variable from one cycle to another ranging from 5 to 30 ml. Middle panels: Variability was also observed in measurements of pressure: (C) mean airway pressure, (D) peak inspiratory pressure. Bottom panels: (E) After starting NAVA, the respiratory rate became very variable over time. (F) Edi max corresponds to the peak of electrical activity of the diaphragm. The highest Edi values (recorded between 19:00 and 22:00) drove assistance with the highest pressures. A decrease in signal intensity was accompanied by a decrease in pressure, corresponding to an improvement in lung function. The requirement for oxygen decreased at the same time. Abbreviations: FiO 2 , fraction of inspired oxygen; Edi, electrical activity of the diaphragm; NAVA, neurally adjusted ventilatory assist. proportion to the measured electrical activity of the diaphragm. This helps to limit the periods of insufficient assist delivery that could induce respiratory muscle fatigue with increased oxygen consumption, and periods of overassistance that can generate intrinsic PEEP with an inadequate increase in intrathoracic pressure [13] . NAVA can also prevent air swallowing and gastric distension by optimization of patient-ventilator synchrony [14] . Third, it is likely that NAVA can help clinician avoiding inappropriate ventilator settings that overload (or underload) respiratory muscles, preventing recovery. Finally, improvements in pulmonary gas exchange, systemic blood flow and oxygen supply to tissues which have been observed when spontaneous breathing has been maintained during mechanical ventilation with clinical improvement in the patient's condition [15] , are assumed to occur in NAVA, when breathing efforts by the patient and the initiated breaths are in synchrony. Cost effectiveness studies are required, as NAVA requires probes that are single-patient use. It is possible that improving comfort provided by better synchronization between spontaneous breathing and mechanical ventilation could reduce the sedation and thus shorten duration of ventilation. Based on three individual cases, NAVA appears to be a useful mode for weaning from a respirator and is an effective alternative treatment for children with severe respiratory distress. NAVA provides a respiratory support that is in harmony with the spontaneous efforts of breathing, allowing a decrease in inspiratory pressures and oxygen needs. Larger studies are required to compare NAVA with conventional respiratory support in children with various etiologies of respiratory distress.

Evaluation of Approaches to Identify the Targets of Cellular Immunity on a Proteome-Wide Scale

BACKGROUND: Vaccine development against malaria and other complex diseases remains a challenge for the scientific community. The recent elucidation of the genome, proteome and transcriptome of many of these complex pathogens provides the basis for rational vaccine design by identifying, on a proteome-wide scale, novel target antigens that are recognized by T cells and antibodies from exposed individuals. However, there is currently no algorithm to effectively identify important target antigens from genome sequence data; this is especially challenging for T cell targets. Furthermore, for some of these pathogens, such as Plasmodium, protein expression using conventional platforms has been problematic but cell-free in vitro transcription translation (IVTT) strategies have recently proved successful. Herein, we report a novel approach for proteome-wide scale identification of the antigenic targets of T cell responses using IVTT products. PRINCIPAL FINDINGS: We conducted a series of in vitro and in vivo experiments using IVTT proteins either unpurified, absorbed to carboxylated polybeads, or affinity purified through nickel resin or magnetic beads. In vitro studies in humans using CMV, EBV, and Influenza A virus proteins showed antigen-specific cytokine production in ELIspot and Cytometric Bead Array assays with cells stimulated with purified or unpurified IVTT antigens. In vitro and in vivo studies in mice immunized with the Plasmodium yoelii circumsporozoite DNA vaccine with or without IVTT protein boost showed antigen-specific cytokine production using purified IVTT antigens only. Overall, the nickel resin method of IVTT antigen purification proved optimal in both human and murine systems. CONCLUSIONS: This work provides proof of concept for the potential of high-throughput approaches to identify T cell targets of complex parasitic, viral or bacterial pathogens from genomic sequence data, for rational vaccine development against emerging and re-emerging diseases that pose a threat to public health.

The development of high throughput techniques to identify the targets of cellular immunity on a proteome-wide scale will facilitate the development of vaccines against complex diseases. Almost all vaccines currently licensed for human use rely on antibody responses against the target pathogen. None are designed to induce protective T cell response, yet T cell responses are implicated as critical in protection against many pathogens, especially those with an intracellular stage such as the causative agents of malaria, leishmaniasis and Chagas disease [1, 2] . CD4 + T cells also play a key role in enhancing the pathogen-specific antibody responses [3] . The elucidation of the genome, proteome and transcriptome of important human pathogens, including the Plasmodium parasite, has provided a wealth of data that can potentially be mined to identify, on a proteome-wide scale, novel target antigens recognized by T cells and antibodies. However, how to effectively mine this data has proved challenging. In particular, technologies such as conventional protein expression methodologies which are well established on an individual antigen basis often cannot be translated directly to a whole proteome scale. An important achievement of the scientific community, therefore, has been the development of technologies that allow the high throughput expression of recombinant proteins. These In Vitro Transcription and Translation systems (IVTT) or cell-free systems offer several advantages over traditional cell-based expression methods and are suitable for high throughput strategies due to reduced reaction volumes and process time [4, 5, 6] . Additional advantages include easy modification of reaction conditions for improving production of complex proteins, decreased sensitivity to product toxicity, and high yield. Importantly, cell-free systems have proved capable of generating proteins from complex parasites that have been difficult to produce in traditional cell-based systems, such as Plasmodium proteins [7, 8, 9, 10] . The most efficient expression to date has been achieved with an E. coli cell-free system which has yielded more than 93% efficiency of expression with a panel of 250 P. falciparum (Pf) proteins [11] . Eukaryotic based cell-free systems are also available with wheat germ, rabbit reticulocytes and insect cells. The eukaryotic lysate is considered by some to provide a better platform for production of complex proteins particularly with regard to post-translational modifications. Recently, up to 75% efficiency of production of Pf proteins in the wheat germ system has been reported [12, 13] but this is still less efficient than that observed with the E. coli system [11] . The combination of these tools with large-scale cloning strategies, such as recombinatorial cloning, allow the generation of complete proteomes in vitro for multiple purposes. Several reports have shown the application of these tools for identifying antibody targets of complex diseases using protein arrays [11, 14, 15] . Such studies have established the feasibility of identifying, from a set of thousands of antigens, the most immunogenic targets of antibody responses which may correlate with protection as indicated by clinical disease stage classification [16] or virus neutralising activity [17] . More challenging is the high throughput elucidation of T cell targets which is clearly of importance for those diseases where cell mediated immunity is implicated in protection, as well as for antibody mediated immunity where T cell help would be beneficial. Recently, the use of the E. coli based cell-free system (Rapid Translational system, RTS) for profiling of CD4 + T cell responses to vaccinia virus in humans was reported [18] . In that study, 180 predicted open reading frames of the vaccinia genome were expressed in the RTS system and unpurified RTS reaction products were tested for recognition by vaccinia virus-enriched T cell lines derived from 11 Dryvax smallpox vaccines, using 3 Hthymidine proliferation assays. Another study reported the use of IVTT products affinity purified on protein G-conjugated carboxylate microsphere beads to stimulate proliferative responses of polyclonal short-term T cell lines from cattle immunised with purified A. marginale outer membranes [19] . Of note, both studies used T cell lines, rather than unpurified splenocytes or bulk peripheral blood mononuclear cells, and neither reported a systematic evaluation of the T cell screening process. Herein, we present a series of in vitro and in vivo experiments designed to demonstrate proof-of-concept for high throughput identification of antigens recognised by T cell responses in human or murine systems, using IVTT products unpurified, affinity purified through nickel resin or magnetic beads, or absorbed in beads to enhance the cell mediated immunogenicity by promoting dendritic cell uptake [20, 21] . IVTT produced antigens of FluM and FluHA from Influenza A virus [22] , CMVpp65 from Cytomegalovirus [23] and EBNA3A from Epstein-Barr virus [24] were assayed using bulk human PBMC for T cell recognition. Additionally, Plasmodium yoelii circumsporozoite protein (PyCSP) [25] IVTT products were assayed for antigenicity in vitro using splenocytes from PyCSP-immunized mice, and for immunogenicity in vivo as assessed by capacity to boost a PyCSP-specific immune response primed by plasmid DNA. In both human and murine systems, T cell responses were evaluated by IFN-c ELISpot and Cytometric Bead Array (CBA) cytokine assays. Robust IFN-c, TNF-a and IL-10 responses were detected, and IVTT products affinity purified through nickel resin or magnetic beads were highly effective. Considering the loss associated with purification, the nickel resin method proved the most optimal. Coding sequences were expressed by coupled transcriptiontranslation in the E. coli cell-free IVTT system. An aliquot of each reaction mixture was analysed by SDS-PAGE and western blot, followed by chemiluminescence of the membrane to visualize the products formed. All four viral (FluHA, FluM, CMVpp65, EBNA3) and one parasite (PyCSP) recombinants could be produced using the manufacturer's recommended conditions. The average yield of each full-length recombinant, detected using an anti-HA antibody directed against the C-Terminal HA tag, was as follow: FluM, 1.15 mg/ml; FluHA, 0.73 mg/ml (two moles of HA tag per molecule of FLU-HA); CMVpp65, 0.7 mg/ml; EBNA3A, 4 ng/ml; and PyCSP, 0.6 mg/ml ( Figures 1A and B and Figure S1A ). Western blot analysis using an antibody directed against the N-Terminal 6xHIS tag identified the presence of partial products for each viral recombinant (PyCSP not tested) ( Figure S1B ). Solubility analysis showed that a high proportion (at least 70%) of each recombinant was insoluble. A series of optimization studies including kinetics of expression (3 hr to 6 hr), reaction temperature (16uC, 25uC, 30uC), speed (stationary, 300 rpm, 600 rpm), addition of protease inhibitor cocktails, or addition of non-ionic detergents to promote protein solubilisation (Triton X-110 or Figure 1 . Recombinants produced using E. coli cell-free IVTT system. Western Blot and quantification of (A) viral antigens FluM, FluHA, CMVpp65 or EBNA3A and (B) parasite antigen PyCSP pIVEX HisHA IVTT products, probed with mAb against the C-terminal HA tag. Whole IVTT extracts (5 ml) of each antigen were run on a 12% NUPAGE gel, transferred to a PVDF membrane, and probed with anti-HA HRP antibody (1:500 dilution). Protein expression was quantitated against an IVTT-produced recombinant P. falciparum (PF14_0051) protein of known concentration expressing the same N-terminal 6xHis and C-terminal HA tags. Yields for each full length antigen were as follow: FluM, 1.15 mg/ ml; FluHA, 0.73 mg/ ml; CMVpp65, 0.7 mg/ ml; EBNA3A, 4 ng/ ml; and PyCSP, 0.6 mg/ ml. doi:10.1371/journal.pone.0027666.g001 Triton X-114) had no significant effect on the yield of full-length protein or partial products, or protein solubility (data not presented). Accordingly, a standard protocol of 4 hrs incubation at 30uC and 300 rpm was adopted for the IVTT reactions. IVTT products were used either unpurified (whole extract), associated to Polybeads or ProteinG beads, or purified using NI-NTA resin or MagneHis Ni-particles. The MagneHis purification method was associated with a very low recovery yield and this yield was much lower than with the NI-NTA system. For example, for PyCSP, the loss associated with purification using Ni-NTA or MagneHis was approximately 30% and 80% respectively. We also evaluated production of the viral antigens in an insect cell based cell-free system (PyCSP not tested). In a limited number of attempts using the manufacturer's recommended conditions, only two of the four viral protein targets (FluM and CMVpp65) could be produced in this eukaryotic system. However, in contrast to results with the E. coli system, solubility analysis showed that at least 90% of the recombinants were soluble and partial polypeptides were not detected ( Figure 2 ). In summary, all FluHA, FluM, CMVpp65, EBNA3 and PyCSP IVTT recombinants could be produced in the E. coli cell-free IVTT system, using the manufacturer's recommended conditions. However, at least 70% of each IVTT product was insoluble. In contrast, only two of the four viral antigens could be produced in the insect cell based system but in this system at least 90% of the product was soluble. In vitro stimulation of PyCSP antigen-specific T cell response using IVTT products The ability of IVTT products to stimulate an antigen-specific T cell response in vitro was evaluated in the P. yoelii CSP model. Splenocytes from mice (n = 5/group) immunized with VR2516 PyCSP plasmid DNA or VR1020 control plasmid were stimulated in vitro with unpurified rPyCSP IVTT; rPyCSP IVTT associated to Polybeads or ProteinG beads; rPyCSP IVTT purified using NI-NTA resin, MagneHis Ni-particles, or anti-HIS; or synthetic peptides representing defined T cell epitopes from PyCSP. Antigen-specific cytokine production was assayed using IFN-c ELISpot or CBA assays. The number of IFN-c SFCs was significantly higher with splenocytes from VR2516 immunized mice as compared to VR1020 immunized mice when stimulated in vitro with PyCSP IVTT purified using NI-NTA resin (p = 0.013), MagneHis Ni-particles (p = 0.015) or PyCSP synthetic peptides (p = 0.003) ( Figure 3A ). There was no significant difference when splenocytes were stimulated with either unpurified IVTT or unpurified IVTT associated to Polybeads or Protein G beads ( Figure 3A ). Unexpectedly, the use of wells precoated with anti-HIS mAb to capture the tagged proteins was poorly effective with no difference noted between the VR2516 and VR1020 groups. Markedly higher background reactivity was noted with the unpurified or bound IVTT preparations as compared to the purified preparations, presumably due to the presence of high levels of LPS and proteins in the E. coli extract ( Figure 3A) . Consistent with the ELISpot data, the amount of secreted IFN-c by CBA in cultures of splenocytes from VR2516 versus VR1020 immunized mice was significantly greater in cultures stimulated in vitro with PyCSP IVTT purified using NI-NTA resin (p = 0.005) or MagneHis Ni-particles (p = 0.003) or PyCSP synthetic peptides (p = 0.02) ( Figure 3B ). There was no significant difference when splenocytes were stimulated with either unpurified IVTT or unpurified IVTT associated to Polybeads or Protein G beads. No significant antigen-specific IL-10, TNF-a or IL-6 responses could be detected upon stimulation with purified or unpurified IVTT products; responses were detected upon stimulation with PyCSP peptides (Figures S2A and S2B ; data not presented for IL-6). In summary, robust antigen-specific IFN-c responses from splenocytes of mice immunized with PyCSP plasmid DNA could be induced following in vitro stimulation with IVTT products purified by either NI-NTA resin or MagneHis Ni-particles, but not by unpurified IVTT reactions or IVTT products associated to Polybeads or Protein beads. There was no significant difference between responses induced by NI-NTA resin and MagneHis Niparticle purifications, but there was a much greater loss of recombinant associated with purification using the MagneHis Niparticles as compared to NI-NTA resin. In general, responses induced by stimulation with the purified PyCSP IVTT products were comparable to responses induced by stimulation with a pool of synthetic peptides representing defined T cell epitopes from PyCSP. In vivo stimulation of PyCSP antigen-specific T cell response using IVTT products We next evaluated the capacity of unpurified and purified IVTT products to boost PyCSP primed immune responses in vivo. Mice (n = 5/group) were primed with VR2516 PyCSP plasmid DNA and then boosted in vivo with unpurified rPyCSP IVTT, PyCSP IVTT associated to Polybeads or ProteinG beads, or PyCSP IVTT purified using NI-NTA resin or MagneHis Niparticles, all adjuvanted with Alum. Splenocytes were stimulated in vitro with A20 target cells transfected with VR2516 PyCSP DNA or A20 cells pulsed with PyCSP synthetic peptides. Antigen-specific cytokine production was assayed using IFN-c ELISpot or CBA assays. IFN-c responses were detected by both ELISpot and CBA following in vivo boosting with PyCSP DNA (p,0.001 for ELISpot and p,0.05 for CBA) or PyCSP IVTT purified using either NI-NTA resin or MagneHis Ni-particles (p,0.01 for ELISpot), as compared to Alum control ( Figures 4A and 4B ). Responses stimulated by purified IVTT products were almost as robust as those stimulated with PyCSP DNA. Low responses were noted when mice were boosted with unpurified whole extract or unpurified IVTT associated to either Polybeads or ProteinG beads. Significant IL-10 responses were also detected following boosting with PyCSP IVTT purified using either NI-NTA resin (p,0.001) or MagneHis Ni-particles (p,0.001), as compared to Alum control, but not following boosting with unpurified whole extract or IVTT products associated to either Polybeads or ProteinG beads, or with PyCSP DNA ( Figure S2C ). TNF-a responses were not significantly different from the controls for any of the conditions ( Figure S2D ). IL-2 responses were induced only by purified IVTT products (data not presented). No significant responses were detected for other cytokines assayed using the CBA assay. In summary, consistent with the results following in vitro stimulation with IVTT products, robust antigen-specific IFN-c responses could be induced by in vivo boosting with PyCSP IVTT products purified by either NI-NTA resin or MagneHis Niparticles, but not by unpurified IVTT reactions or IVTT products associated to Polybeads or ProteinG beads. There was no significant difference between purification with either NI-NTA resin or MagneHis Ni-particle, nor between purified IVTT products and plasmid DNA. The magnitude of Th1 (IFN-c and TNF-a) responses stimulated in vitro by purified IVTT PyCSP products were similar to those stimulated by PyCSP plasmid DNA. In vivo stimulation of PyCSP antigen-specific antibody response using IVTT products The ability of IVTT products to stimulate an antigen-specific antibody response was also determined. Sera from mice (n = 5/group) primed with VR2516 PyCSP plasmid DNA and boosted in vivo with rPyCSP IVTT purified using NI-NTA resin plus Alum were evaluated by ELISA against synthetic peptide representing the recombinant PyCSP protein. Data demonstrated that in vivo boosting with PyCSP IVTT products significant boosted antigen-specific antibody responses relative to responses induced by PyCSP plasmid DNA in the absence of boosting ( Figure 5 ). The antigenicity of IVTT products in human PBMC cultures was next evaluated. The viral antigens FluM, FluHA, CMVpp65 and EBNA3A produced by IVTT were used unpurified, associated to Polybeads or ProteinG beads, or purified using NI-NTA resin or MagneHis Ni-particles to assay recall cytokine responses from PBMCs of 10 healthy humans by IFN-c ELISpot or CBA. Consistent with the results in murine system, significant IFN-c ELISpot responses (p,0.05) were detected with all IVTT antigens purified using NI-NTA resin ( Figure 6A ) and for all antigens purified using MagneHis Ni-particles except FluM ( Figure 6B ) (as compared to PBS); the lack of responses with FluM was attributed to a low yield post MagneHis Ni purification (data not presented). Unexpectedly, significant IFN-c ELISpot responses were also noted with all IVTT viral antigens used unpurified (compared to empty pIVEX HisHA IVTT) ( Figure 6C ) or associated to ProteinG beads (compared to empty pIVEX HisHA IVTT associated with ProteinG beads) ( Figure 6D ), but only for only two or the four IVTT viral antigens associated with Polybeads (FluM and FluHA; compared to empty pIVEX HisHA IVTT associated with Polybeads) ( Figure 6E ). Also consistent with the murine system, no significant responses were detected for any antigens when the ELISpot wells were precoated with anti-His mAb to capture the His-tagged IVTT proteins. The background responses with the unpurified or bound IVTT products were greater than for purified IVTT products, Figure 6 . Antigen-specific IFN-c ELIspot responses by human PBMC stimulated with IVTT-proteins. PBMCs were cultured with IVTTproduced FluM, FluHA, pp65 and EBNA3A purified using (A) NI-NTA nickel resin or (B) MagneHis Ni-particles; (C) unpurified; associated to (D) ProteinG beads or (E) Polybeads; or (F) added to wells precoated with Anti-His. Negative controls were medium only, unpurified whole extract, and whole extract associated to Polybeads or Protein G beads; positive control was CEF peptide pool. IFN-c ELIspot responses (spot forming cells, SFC) of cultured PBMCs were analysed after 36 hrs stimulation. Data are presented as mean +/-standard deviation of 10 volunteers. *P,0.05, **P,0.01 and ***P,0.001 compared to negative controls. doi:10.1371/journal.pone.0027666.g006 presumably as a result of prior exposure of the human subjects to E. coli bacteria or presence of LPS. Significant IFN-c responses to the CEF peptide pool p,0.0001, compared to PBS) were noted for all subjects (data not presented). For all viral antigens, robust IFN-c responses were detected by CBA in cultures stimulated at 1:100 dilution with IVTT products purified using NI-NTA resin ( Figure 7A ) or MagneHis Ni-particles ( Figure 7B ) but not with purified products at 1:1,000 or 1:10,000 dilutions, presumably due to the low amount of antigen in culture ( Figures 7A and 7B ). With unpurified IVTT products or IVTT antigens associated to Polybeads or ProteinG beads, IFN-c responses were generally better at the higher dilution (1:10,000.1:1,000.1:100) where background E. coli responses were lower (Figures 7C-E) . However, none of these IFN-c responses were statistically different from the controls due to variations in the cytokine levels between individuals (p,0.05 only for EBNA3A-Protein G beads at 1:1000). The profile of TNF-a responses by CBA was very similar to that of IFN-c ( Figure S3 ). For all viral antigens (except FluM-Ni-NTA), TNF-a responses were statistically significant compared to controls for cultures stimulated at 1:100 dilution with IVTT products purified using NI-NTA resin ( Figure S3A ) or MagneHis Niparticles ( Figure S3B ). Responses were also significant at 1:1,000 dilutions for CMVpp65 and EBNA3 all purified using NI-NTA resin or MagneHis Ni-particles; and FluHA purified with MagneHis Ni-particles. The TNF-a profile was also similar to that of IFN-c for unpurified and bead-associated IVTT products, with the highest responses almost always detected at 1:10,000 (except for EBNA3) (Figures S3C-E) . However, none of these TNF-a responses were statistically significant (except for unpurified CMVpp65 at 1:10,000), as noted above for IFN-c. The overall level of TNF-a was enhanced by association to Polybeads and Protein G beads, relative to unpurified IVTT products (up to 7-fold with EBNA3A-Polybeads at 1:100 dilution). The profile of IL-10 responses by CBA was very similar to that of IFN-c and TNF-a ( Figure S4 ) for cultures stimulated with IVTT products purified using NI-NTA resin ( Figure S4A ) or MagneHis Ni-particles ( Figure S4B ) with the best responses detected at 1:100 ( Figure S4 ). However, for unpurified or beadassociated IVTT products, the IL-10 profile was inverse to that of IFN-c and TNF-a with the best responses at a 1:100 dilution ( Figures S4C-E) , consistent with potential immunosuppressive effects of IL-10. Significant response was found only in EBNA3A 1:100 purified with MagneHis Ni-particles. As noted for TNF-a, the overall level of IL-10 was enhanced by association to Polybeads and Protein G beads relative to unpurified IVTT products (up to 7-fold with EBNA3A-Polybeads at 1:100 dilution). Responses for other cytokines assayed by CBA were not significant (data not presented). In summary, consistent with the murine data, robust antigenspecific IFN-c and TNF-a responses from human PBMCs could be induced following in vitro stimulation with IVTT products purified by either NI-NTA resin or MagneHis Ni-particles. In contrast to the murine system, positive IFN-c and TNF-a responses to some antigens could also be induced by unpurified IVTT reactions or IVTT products associated to Polybeads or Protein beads, indicating that the E. coli and LPS background of the whole extract was not sufficient to mask the immunogenicity generated by the IVTT recombinant products. For some but not all of the evaluated antigens, association of IVTT products to Polybeads or Protein G beads enhanced the antigenicity about 2fold to 4-fold. The rapidly growing amount of genetic and proteomic information available in the post-genomic era precludes the use of traditional cell-based protein expression systems to screen proteins of interest for their potential as vaccine targets. The development of cell-free based systems, first described in 1960's [26] , allows the high throughput recombinant expression of thousands of proteins in reduced volume and time. The most popular cell-free systems are based in E. coli, wheat germ and rabbit reticulocytes extracts [4, 5, 27] . This platform has been applied in a number of proteomics-based studies including structural and functional proteomics [28, 29, 30] , protein evolution [31] , unnatural amino acids and protein labeling [32, 33] , protein interaction [34] , diagnostics and therapeutics [35, 36] and protein microarrays [15, 37, 38, 39] . The protein microarray platform, exploiting antigen-specific antibodies present in plasma or sera from exposed or immunized animals or humans, has been of particular interest to our laboratory to identify potential target antigens for malaria vaccine development [10, 11, 40] . Although Plasmodium proteins have proven particularly difficult to express using conventional cell-based methods, efficient expression of P. falciparum proteins ($ 93%) has been obtained using the E. coli cell-free system [11] . The wheat germ system has been also used for production of P. falciparum proteins, but with less efficiency (75%) [12] . The success with cell-free protein expression suggest that this system could be also applied to cellular screening, to identify antigenic targets of T cell responses from genomic sequence data. However, this has not yet been adequately explored despite that T cells play a central role in orchestrating acquired immunity against infectious diseases [41, 42] . Both CD8 + and CD4 + T cells can mediate their effector function directly via cytotoxicity or indirectly via cytokines. CD4 + T cells can also provide help for CD8 + T cells or can recruit and activate B cells for antibody secretion. To date, the only reports have been restricted to measurement of proliferative T cell responses in PBMCs of smallpox vaccines or of cattle immunized with a purified Anaplasma marginale outer membrane preparation to a small number of IVTT produced vaccinia virus proteins or outer membrane proteins, respectively [18, 19, 43] . Neither of those studies comprehensively assessed and optimized the application of IVTT products for large-scale or proteome-wide cellular screening. Accordingly, herein, we report proof of concept for the potential of this approach a strategy for proteome-wide identification of antigens targeted by cell mediated immunity in both viral and parasite models, using IVTT products with specimens from human and mice. Our human studies used well characterized antigens from EBV, CMV and Influenza A virus. which are known to be targeted by both CD4 + and CD8 + T cell responses and with defined T cell epitopes [44] . The murine studies used the well characterized sporozoite coat protein, the circumsporozoite protein (CSP), from the P. yoelii rodent malaria parasite. We tested each IVTT antigen individually and presented to T cells in different forms -either unpurified, purified using NI-NTA resin or MagneHis Niparticles, or associated to Polybeads or ProteinG beads. Synthetic peptides representing defined T cell epitopes from the respective antigens were assayed in parallel as positive controls. The primary immune readout was IFN-c production due to its crucial role in protection or pathogenesis of complex diseases [45] . For all four viral and one parasite antigens tested in this study, partial products were common. The production of partial fragments in E. coli based IVTT reactions has been described previously and attributed to a phenomena called translational pausing but the presence of partial products did not adversely affect immunogenicity or antigenicity [46] . Since the antigens produced by E. coli based IVTT were at least partially insoluble, simple techniques for purification, such as antibody coated beads, were not effective. We therefore explored alternative purification options including Ni-NTA affinity resin and MagneHis Niparticles. Both purification systems tested using denaturating conditions proved efficient but a much higher product recovery was obtained with the Ni-NTA resin as compared to the MagneHis Ni-particles. Unexpectedly, with the eukaryotic based insect cell-free system, the solubility of the recombinants increased to almost 100%, but only two of the four proteins expressed in the E. coli cell-free system could be expressed in the insect cell based system, at least for the limited number of times expression was attempted. These data suggest that further studies with the insect cell-free system are warranted. Newer technologies of high throughput purification such as affinity ZipTips (Millipore, Ireland) and Ni-NTA plates (Qiagen, Valencia, CA) can facilitate the purification process, but do not effectively deal with solubility issues. In both human/virus and murine/parasite models, IVTT products purified by either NI-NTA resin or MagneHis Niparticles were highly effective inducers of antigen-specific IFN-c and TNF-a responses in vitro. In the human/virus but not murine/ parasite model, antigen-specific IFN-c and TNF-a responses could also be induced by unpurified IVTT reactions or IVTT products associated to Polybeads or Protein beads, indicating that the E. coli and LPS background of the whole extract was not sufficient to mask the immunogenicity of the IVTT recombinant products. Addition of carboxylate beads to unpurified IVTT products increased the immunogenicity of the IVTT viral products by up 1.8-fold for IFN-c, 2.6-fold for IL-10 and 7-fold for TNF-a. The most robust and specific antigen-specific IFN-c and TNF-a responses (by CBA) were detected at the highest dilution of IVTT product, where background E. coli responses were lowest. This is a favorable scenario for proteomic-wide scale cellular screening, as the use of highly diluted IVTT products is more cost-effective. Unexpectedly, poor results were obtained with ELIspot wells precoated with anti-IFN-c mAb as well as anti-HIS mAb to bind the HIS tag on the IVTT product. In vivo studies with PyCSP IVTT products confirmed that the target protein was produced and that the IVTT produced proteins were immunogenic. These data demonstrate the potential of IVTT products as a useful tool for the proteome-wide screening of cellular targets of viral, parasitic or bacterial immunity Overall, IVTT products affinity purified through nickel resin or magnetic beads proved the most efficient inducers of sensitive and specific antigen-specific cytokine responses, the nickel resin method was associated with the greater yield post-purification. Although not specifically evaluated herein, it is likely that such cell-free approaches may be suited to the identification of targets of CD4 + T cell responses, but not targets of CD8 + T cell responses due to a requirement for target antigen processing and presentation [43, 47, 48] . Rather, epitopebased approaches based on prediction of high affinity binding class I T cell epitopes using computerized algorithms, such as that reported by us previously [49] are probably more appropriate. Overall, the work reported here provides proof of concept for the potential for high-throughput identification from genomic sequence data of antigenic targets of T cell responses from complex pathogens which threaten public health. Such antigens may represent promising candidates for the development of vaccines that have thus far proved elusive. Ten healthy adult Caucasian volunteers (five male and five female; mean age 36.367.1 years old) were recruited with written informed consent under a protocol (P1111) approved by the Queensland Institute of Medical Research Human Research Ethics Committee. All studies with human specimens were approved by the Queensland Institute of Medical Research Human Research Ethics Committee (P1111) and conducted in compliance with all applicable regulations governing protection of human subjects. Although the history of vaccination in these subjects has not been documented, it is likely that all would have been vaccinated against or exposed to influenza, EBV, and CMV, given the documented prevalence of these viruses in human populations; and all were known to respond to the CEF peptide pool which comprises CD8 + T cell epitopes from FLU, EBV and CMV. Female BALB/c (H-2 d ) mice aged 6-8 weeks were purchased from The Animal Resource Centre (Perth, WA) and maintained under standard conditions. All studies were approved by the Queensland Institute of Medical Research Animal Ethics Committee (protocol P1111). The plasmid DNA vaccines encoding the P. yoelii circumsporozoite protein (CSP; VR2516) and the control plasmid (VR1020) have been previously described [50, 51] . The DNA vaccines were prepared using the EndoFree Plasmid Mega Kit (Qiagen, Valencia, CA) according to manufacturer's instructions and administered in sterile saline. A custom vector was developed for protein expression in the E. coli cell-free transcription translation system. This vector, called pIVEX HisHA AmpR, was modified from the commercially available pIVEX 2.4d and pIVEX 2.5d vectors (Roche Applied Science, Mannheim, Germany) by incorporating both the N-terminal HIS tag and C-terminal HA tag into the one vector backbone. The HA tag was released from pIVEX 2.5d by digestion with BamHI and XmaI restriction enzymes (New England Biolabs, Ipswich, MA) for 4 hrs at 37uC (1 mg DNA/1 unit enzyme, 20 ml volume) and the fragment then extracted from a 3% agarose gel and purified using a commercially available QIAquick PCR purification kit (Qiagen, Germany) according to manufacturer's instructions. The purified HA-tag was ligated overnight at 16uC into BamHI and XmaI digested pIVEX 2.4d vector using 400 units T4 ligase (New England Biolabs), 25 ng of linearized pIVEX 2.4d vector and 70 ng of the HA-tag insert in a 20 ml reaction. The ligated product was purified from a 1% agarose gel as described above, transformed into TOP10 thermocompetent cells (Invitrogen), and grown on LB plates in the presence of 100 mg/ml ampicillin. Positive colonies were screened by colony PCR using 0.2 units/ml Expand Taq polymerase in Buffer 2 (Roche Diagnostics), 0.4 mM dNTPs and 0.4 mM each primer (forward: TAATACGACTCACTATAGGG; reverse TGCTAGT-TATTGCTCAGCGG) using the following conditions: initial denaturation at 95uC for 5 min; 35 cycles at 95uC for 30 sec, 55uC for 30 sec and 68uC for 1 min; and a final extension at 68uC for 10 min. #NC_007605) and P. yoelii circumsporozoite protein (PyCSP; strain 17XNL, accession #J02695) were amplified by PCR using gene-specific primers flanked with restriction enzyme sites ( Table 1 ). The 50 ml PCR reaction contained 0.2 units/ml Expand Taq polymerase in Buffer 2 (Roche diagnostics), 0.4 mM dNTPs and 0.4 mM each primer and 50 ng DNA template. PCR conditions were: initial denaturation at 95uC for 5 min; 45 cycles at 95uC for 30 sec, 55uC for 30 sec and 68uC for 3 min; and a final extension at 68uC for 10 min. The fragments corresponding to the expected size were excised from a 1% agarose gel, purified using the QIAquick PCR purification kit (Qiagen, Germany), and digested overnight at 37uC using the restrictions enzymes for which the cutting site was present in the flanking sequence. The pIVEX HisHA vector was also digested overnight at 37uC with the same set of restriction enzymes. After cleanup using the PCR purification kit, the gene fragments and the linearized vector were ligated overnight at 16uC at an insert:vector ratio of 1:20 using 30 ng of vector and 400 units T4 ligase (New England Biolabs) in a 20 ml reaction volume. The ligation product was transformed into TOP10 thermo-competent cells, grown overnight at 37uC on LB ampicillin plates (100 mg/ml), and colonies screened by PCR using the same protocol for the customized vector, except that the extension time of the colony PCR was increased to 3 min. One positive colony of each construct was grown overnight at 37uC in LB containing ampicillin (100 mg/ml) and the plasmid purified using the QIAprep Spin Miniprep kit (Qiagen) according to manufacturer's instructions. In a few instances where the IVTT production was poor, the DNA was further purified using phenol/chloroform extraction and ethanol precipitation which improved the yield, but this was not routinely done. All constructs were confirmed by DNA sequencing before use in the IVTT reactions. Recombinant proteins were synthesized by cell-free in vitro transcription and translation using the Rapid Translation System 100 E. coli HY kit (Roche Diagnostics) or EasyXpress Insect Cell Protein kit (Qiagen) according to the manufacturer's instructions. In some experiments, protease inhibitor cocktails (Cat P2714, Sigma-Aldrich, St Louis, MO; Cat 1836170, Roche Diagnostics), or addition of non-ionic detergents (Triton X-110 or Triton X-114) were added according to manufacturer's recommended protocol: 4 hr at 25uC. The recombinant antigens were stored at 220uC and used as immunogens for in vitro and in vivo up to one week after production. Protein expression of the full-length product (as evidenced by immunoblot using the anti-HA Cterminal tag mAb) was quantitated against an IVTT-produced recombinant P. falciparum (PF14_0051) protein of known concentration expressing the same N-terminal 6xHis and C-terminal HA tags. Immunoblot images were acquired in TIF format and analysed using the AnalySIS LS software (version 5.0; Soft Imaging Systems GmbH, Germany). The E. coli IVTT products were purified using two different methods based on affinity to the N-terminal 6xHIS tag. One method used mini-spin columns with a cellulose acetate filter (Pierce, Rockford, IL) and NI-NTA resin (Qiagen). For that, 200 ml of NI-NTA resin was added to a spin column connected to a collection tube. The column was washed twice with 600 ml of milliQ water and twice with 600 ml of Binding Buffer (50 mM NaH 2 PO 4 , 300 mM NaCl, 20 mM Imidazole and 8 M Urea pH 8) using centrifugation at 1500 rpm for 2 min. Then 50 ml of IVTT reaction was mixed with 700 ml of Binding Buffer and incubated at 4uC for 20 min in an orbital mixer, to solubilize the inclusion bodies. The solubilized sample was added to a NI-NTA resin column and incubated at 4uC for 30 min in an orbital mixer to maximize protein binding to the resin. The flow-through was removed via centrifugation at 1500 rpm for 2 mins and the column washed three times with 700 ml of Binding Buffer. The recombinant was eluted with 200 ml of Elution Buffer (50 mM NaH 2 PO 4 , 300 mM NaCl, 250 mM Imidazole and 8 M Urea pH 8) incubated at 4uC for 20 min in an orbital mixer before collection via centrifugation. The second method of purification used the commercially available MagneHis Protein Purification System (Promega, Madison, WI) according to the manufacturer's instructions. For that, 50 ml of IVTT reaction was mixed with 700 ml of Binding Buffer (100 mM HEPES, 20 mM Imidazole and 8 M Urea pH 7.5) and incubated at 4uC for 20 min in an orbital mixer, to solubilize the inclusion bodies. Then, 30 ml of MagneHis Niparticles was added to the sample and the mixture incubated at room temperature for 2 min. The sample tubes were connected to the MagneSphere Magnetic Separation Stand (Promega) and, after 30 sec of binding, the flow through was removed and discarded. The magnetic beads with bound protein were washed 3 times with 700 ml of Binding Buffer and the recombinant eluted in 200 ml of Elution Buffer (100 mM HEPES, 500 mM Imidazole and 8 M Urea pH 7.5) incubated at 4uC for 20 min in an orbital mixer before collection. The eluted recombinants from both methods were dialyzed against PBS (desalting step) using Amicon ultra 0.5 ml 3 kDa cutoff centrifugal filters (Millipore, Ireland). For that, the centrifugal filters were wet with 300 ml of PBS pH 7.0 before adding 200 ml of the eluted protein. After 30 min of centrifugation at 14,000 rpm at 4uC, the volume of the protein was reduced to approximately 100 ml and 400 ml of PBS pH 7.0 were added. This process was repeated four times and the resultant sample collected and stored at 220uC. Protein expression was confirmed by western blot. IVTT products were diluted 1:1 in 2x reducing sample buffer (125 mM Tris pH 6.8, 4% SDS, 10% glycerol, 0.006% bromophenol blue, 2% beta-mercaptoethanol) and denatured at 95uC for 5 mins. Samples (2.5 ml of E. coli IVTT whole extract or supernatant, and 5 ml of insect cell IVTT whole extract or supernatant) were run in parallel with molecular weight standard (BenchMark pre-stained protein marker, Invitrogen) on a 4-12% NuPage SDS Page gel (Invitrogen, Carlsbad, CA, USA) at 120 V for 70 min. Samples were transferred onto a PVDF membrane (Bio-Rad, Hercules, CA) pre-wet with 100% methanol, using the BioRad Mini Trans-Blot system at 100 V for 1 hr, according to manufacturer's instructions. Membranes were blocked with PBS containing 5% skim milk powder overnight at 4uC with shaking, washed 3 times with PBS containing 0.05% Tween20, then probed with anti-HIS HRP (1:5000; Roche Diagnostics) or anti-HA HRP (1:500) (Roche Diagnostics) for 1 hr at room temperature. Membranes were Immunization of mice using the VR2516 PyCSP DNA vaccine and PyCSP IVTT product Mice (n = 5/group) were immunized intramuscularly in the tibialis anterior muscle with 50 mg of VR2516 PyCSP DNA vaccine or VR1020 negative control plasmid, two times at 3 week intervals. Splenocytes were harvested from DNA immunized mice or naïve control mice (n = 5) at 3 weeks post-boost for in vitro T cell assays using IVTT products. Serum was collected from select groups for antibody assays. The potential toxicity of IVTT products was assessed prior to use for in vitro stimulation of T cell responses by culturing splenocytes (5610 5 cells/well) of naïve BALB/c mice with IVTT-produced PyCSP either unpurified or purified using NI-NTA resin (Qiagen) or MagneHis particles (Promega) at dilutions of 1:100, 1:1000 and 1:10000, or media alone, and monitoring cell growth for 48 hours. No toxicity due to the unpurified or purified IVTT preparations was apparent, as indicated by cell death. For the in vivo evaluation of IVTT products, mice (n = 5/group) were immunized intramuscularly with 50 mg/100 ml of VR2516 PyCSP DNA vaccine (split between two legs) and boosted 3 weeks later with PyCSP IVTT products formulated with 100 ml Aluminium Hydroxide gel adjuvant (Brenntag Biosector, Frederikssund, Denmark) as follows: A) 15 ml of IVTT reaction (9 mg PyCSP/mouse); B) 15 ml of IVTT reaction purified through NI-NTA resin (6.6 mg PyCSP/mouse); C) 15 ml of IVTT reaction purified through MagneHis Protein Purification System (3 mg PyCSP/mouse); D) 15 ml of IVTT reaction absorbed in 10 ml of PolybeadH Poly(methyl methacrylate) microspheres (PMMA microspheres, 0.08-0.09 mm diameter) (Cat #23570, Polysciences, Inc., Warrington, PA, USA) (9 mg PyCSP/mouse); and E) 15 ml of IVTT reaction associated to 10 ml of Protein G conjugated microspheres/carboxylate beads (Cat #21106, Polyscience) (9 mg PyCSP/mouse). The beads used in the immunizations were washed 3 times with PBS pH 7.0 before addition of the IVTT product, and used without further processing. Positive and negative controls groups (n = 5) were administered 50 mg of VR2516 PyCSP DNA vaccine intramuscularly or 100 ml of Alum mixed with 100 ml of PBS pH 7.0, respectively. Three weeks after immunization mice were sacrificed and the spleens harvested for use as effectors in T cell assays. Spleens were macerated, washed in Dulbecco's solution containing 2% FCS, the red blood cells lysed with 0.09% NH 4 Cl for 5 min at 37uC, and then washed again. The splenocytes were resuspended in complete DMEM containing 10% FCS, 100 U/ml of Penicillin, 100 mg/ml of Streptomycin, 2 mM L-Glutamine and 0.05 mM ß-mercaptoethanol, and cultured in 96-wells flat-bottomed plates for ELIspot or CBA at a concentration of 5610 5 cells/well. For in vitro evaluation of IVTT products, splenocytes were stimulated with unpurified rPyCSP IVTT; rPyCSP IVTT associated to Polybeads or ProteinG beads; or rPyCSP IVTT purified using NI-NTA resin or MagneHis Ni-particles. Unpurified IVTT products were used at a dilution of 1:200 or 1:1000 and purified IVTT products were used at a dilution of 1:100 or 1:200 for evaluation of cytokines in the supernatant using the Cytometric Bead Array assay or IFN-c secreting cells via ELIspot, respectively. Pilot experiments were carried out to find out the optimal dilutions of IVTT products for these assays (data not shown). An IVTT reaction using the empty pIVEX HisHA vector was used as negative control. A pool of synthetic peptides representing defined CD4 + and CD8 + T cell epitopes from PyCSP (residues 57-70, sequence KIYNRNIVNRLLGD [52] ; residues 280-288, sequence SYVPSAEQI [53] ; and resides 280-295, sequence SYVPSAEQILEFVKQI [54] ) at 10 mg/ml of each peptide was used as a positive control. ConA (Sigma-Aldrich, St. Louis, MO) at 5 mg/ml was used as a mitogen control. All stimuli were added in association with 1610 5 cells/well of A20/2 J cells (ATCC clone HB-98) irradiated at 1600 rads, as antigen presenting cells. For in vivo evaluation of IVTT products, splenocytes were stimulated with 2610 4 cells/well of A20 cells transfected with VR2516 PyCSP DNA or VR1020 negative control using the AMAXA Nucleofector transfection kit (AMAXA, Cologne, Germany) according to manufacturer's manual, or with 1610 5 cells/well of A20 cells pulsed with the PyCSP peptide pool. A20 cells alone and medium were used as negative controls and ConA used as mitogen control. AMAXA transfections included parallel reactions with GFP plasmid as a positive control (data not presented); additionally, our laboratory routinely uses AMAXAtransfected A20 cells as APCs for in vitro T cell assays [55, 56] . Peripheral blood mononuclear cells (PBMC) were isolated using standard Ficoll density gradient centrifugation and resupsended in complete RPMI containing 10% human AB serum, 100 units/ml of Penicillin, 100 mg/ml of Streptomycin, 2 mM L-Glutamine, 1 mM Sodium Pyruvate and 25 mM Hepes. Cells were cultured at a concentration of 1610 5 cells/well in 96-wells round-bottomed plates at 37uC in an atmosphere of 5% CO 2 . The cells were stimulated with IVTT-produced FluM, FluHA, CMVpp65 and EBNA3A antigens, either unpurified; associated to Polybeads or ProteinG beads; or purified using NI-NTA resin or MagneHis Niparticles. Unpurified and purified IVTT products were used at a dilution of 1:100, 1:1000 or 1:10000 for evaluation of cytokines in the supernatant using the Cytometric Bead Array assay, or at a dilution of 1:1000 (unpurified) or 1:100 (purified) for evaluation of IFN-c secreting cells via ELISpot. Negative controls included empty pIVEX HisHA IVTT reaction alone or associated to Polybeads or ProteinG beads, or medium only. A CEF peptide pool consisting of 32 synthetic peptides representing defined CD8 + T cell epitopes from human Cytomegalovirus, Epstein-Barr Virus and Influenza Virus (Anaspec, San Jose, CA) at 5 mg/ml total was used as a positive control [44] . Phytohemagglutinin (PHA; Sigma-Aldrich, St. Louis, MO) at 2 mg/ml was used as a mitogen control. IFN-c secreting T cells were enumerated by ELISpot. Briefly, MultiScreen HTS IP 96 plates (Cat MSIPS4510, Millipore, Ireland) were pre-wet with 15 ml/well of 35% ethanol and washed twice with PBS (pH 7.4). For mouse ELIspot assays, wells were coated with 75 ml/well of sterile PBS (pH 7.4) containing 10 mg/ ml anti-mouse IFN-c (Cat 551216, Clone R4-6A2, BD Pharmingen, CA) with or without 1:500 anti-His mAb (Cat H1029, Sigma-Aldrich, St. Louis, MO) overnight at room temperature. For human ELIspot assays, wells were coated with 75 ml/well of carbonate buffer (pH 9.6) containing 10 mg/ml anti-human IFN-c (Cat 3420-3, Clone 1-D1K, MABTECH, Sweden) with or without 1:500 anti-His mAb (Cat H1029, Sigma-Aldrich, St. Louis, MO) overnight at 4uC. Plates were washed twice with 200 ml/well DMEM (splenocytes) or RPMI (PBMC) and blocked with 200 ml/ well of medium containing 10% FCS for 3 hrs at 37uC. After blocking, the wells coated with anti-His mAb were incubated with PyCSP, FluM, FluHA, CMVpp65, EBNA3A or empty pIVEX HisHA IVTT products diluted 1:1000 in DMEM or RPMI containing 2% FCS in a volume of 50 ml/well, overnight at 4uC (to absorb the His-tagged IVTT products to the anti-His mAb coated wells). Plates were washed 3 times with DMEM or RPMI, and the cells added to the IFN-c and His coated wells in triplicates. Alternatively, in the absence of anti-His pre-coating, IVTT products unpurified, associated to Polybeads or ProteinG beads, or purified using NI-NTA resin or MagneHis Ni-particles (see above), were added to the IFN-c coated wells together with the cells (100,000 PBMCs or 500,000 splenocytes), in triplicate. After 36 hrs incubation, plates were flicked to remove the cells and washed 6 times with PBS-Tween 0.05% pH 7.4 (PBS-T). Then 75 ml/well of biotinylated anti-mouse IFN-c (Cat 554410, Clone XMG1.2, BD Pharmingen) at 2 mg/ml in PBS or 75 ml of biotinylated anti-human IFN-c (Cat 3420-6, Clone 7-B6-1, MABTECH, Sweden) diluted 1:1000 in PBS was added to each well. Plates were incubated for 3 hrs at room temperature, washed 3 times with PBS-T, and 75 ml/well of Streptavidin-HRP for splenocytes cultures or Streptavidin-AP (Bio-Rad, Hercules, CA) for PBMC cultures diluted 1:1000 in PBS was added. After 1 hour incubation at room temperature, plates were washed 3 times with PBS-T followed by 3 times with PBS alone, and developed using the AEC substrate set (BD Pharmingen) for mouse splenocytes or SigmaFast BCIP/NBT (Sigma-Aldrich) for humans PBMC cultures, according to manufacturer's instructions. After 10 mins, the plates were rinsed extensively with dH2O to stop the enzymatic reaction, dried and stored in the dark. The number of IFN-c secreting cells was determined using the automated ELISpot reader (AID iSpot Reader, Autoimmun Diagnostika GmbH, Strassberg, Germany). To determine the cytokines secreted by the stimulated splenocytes or PBMC, cells were cultured at a concentration of 5610 5 cells/well in 96-wells flat-bottomed plates (Corning Incorporated, Corning, NY, USA). The supernatant was collected after 48 hrs and 72 hrs stimulation and analyzed with Cytometric Beads Array flex kits (BD Biosciences, San Jose, CA, USA) containing IL-1ß, IL-2, IL-4, IL-5, IL-6, IL-10, IL-12p70, IL-13, IFN-c and TNF-a for mouse and IL-10, IFN-c and TNF-a for human samples. Data were acquired and analyzed using BD FACS Array Bioanalyser and FlowJo 7.6 software. PyCSP-specific antibody responses in mouse sera were evaluated by peptide ELISA, as previously described [57] . Briefly, flatbottomed 96-well microtiter plates (Immulon 4; Dynex Technology Inc., Chantilly, VA, USA) were coated with 100 ml/well of recombinant PyCSP protein at a concentration of 1 mg/ml in carbonate-bicarbonate coating buffer pH 7.4, and incubated overnight at 4uC. Wells were blocked for 1 hr with 2% BSA in PBS containing 0.05% Tween 20 (Blocking Buffer) and washed three times with PBS containing 0.05% Tween 20 (PBS-T). Consecutive dilutions of individual sera diluted in PBS containing 0.01% Tween-20 were incubated for 2 hrs at room temperature. The plates were washed 3 times, and incubated with 100 ml/well Biotinylated anti-mouse IgG (Jackson ImmunoResearch) at a dilution of 1:20,000 for 1 hr. The plates were washed three times and incubated with Streptavidin HRP (BD Biosciences) at a dilution of 1:1000 for 1 hr. The plates were washed and developed for 10 mins with 50 ml/well TMB substrate (Sigma). Reactions were terminated by adding 50 ml of stopping buffer and the OD450 recorded using a VersaMax microplate reader (Molecular Devices, Sunnyvale, CA, USA). Results are expressed as mean OD readings of triplicate wells +/-SE. Data from the in vitro and in vivo tests in mice were analyzed by the Student t-test and two-way ANOVA, respectively, where a p value,0.05 was considered significant. Data from human PBMC ELIspot and CBA were analyzed using two-way ANOVA and one-way ANOVA, respectively. A p value,0.05 was considered significant. Figure S1 Recombinants produced using E. coli cellfree IVTT system. Western Blot of (A) viral antigens FluM (32 kDa), FluHA (68 kDa), CMVpp65 (68 kDa) and EBNA3A (108 kDa), and parasite protein PyCSP (44 kDa), probed with anti-HA antibody, and (B) viral antigens FluM, FluHA, CMVpp65 and EBNA3A probed with anti-His antibody. Whole IVTT extracts (5 ml) of each antigen were run on a 12% NUPAGE gel, transferred to a PVDF membrane, and probed with anti-HA HRP antibody (1:500 dilution) or anti-His HRP antibody (1:5000 dilution). The western probed with anti-HA also detected a cross reactive band of 56 kDa in all expression extracts. The western probed with anti-His mAb showed the presence of partial products probably due to early termination of translation. (TIF) Figure S2 Antigen-specific TNF-a and IL-10 responses of mouse splenocytes stimulated in vivo or in vitro with IVTT-proteins. (A) and (B): splenocytes of mice immunized with VR2516 PyCSP plasmid DNA or VR1020 control DNA were cultured in vitro with unpurified rPyCSP IVTT; rPyCSP IVTT associated to Polybeads or ProteinG beads; rPyCSP IVTT purified using NI-NTA resin, MagneHis Ni-particles, or anti-HIS; or synthetic peptides representing defined T cell epitopes from PyCSP (positive control), as indicated. (C) and (D): splenocytes of mice immunized with VR2516 PyCSP plasmid DNA and boosted in vivo with unpurified rPyCSP IVTT; rPyCSP IVTT associated to Polybeads or ProteinG beads; or rPyCSP IVTT purified using NI-NTA resin or MagneHis Ni-particles; all formulated with Alum adjuvant. Parallel groups of mice were boosted with either Alum only or VR2516 as controls. Splenocytes were cultured in vitro with A20 cells transfected with VR2516 PyCSP plasmid DNA or A20 cells pulsed with synthetic peptides representing defined PyCSP T cell epitopes, as indicated. Secreted TNF-a (A and C) or IL-10 (B and D) in culture supernatant was measured by Cytometric Bead Array (CBA) after 48 hrs stimulation. *P,0.05, **P,0.01 and ***P,0.001 compared to negative controls. (TIF) Figure S3 Antigen-specific TNF-a responses by human PBMC stimulated with IVTT-proteins. PBMCs were cultured with IVTT-produced FluM, FluHA, CMVpp65 and EBNA3A purified using (A) NI-NTA nickel resin or (B) MagneHis Ni-particles; (C) unpurified; associated to (D) ProteinG beads or (E) Polybeads; or (F) added to wells precoated with Anti-His; IVTT products were diluted 1:100, 1:1000, or 1:10,000. Secreted TNF-a in the supernatant of cultured PBMCs was analyzed by Cytometric Bead Array after 48 hrs stimulation. * P,0.05 compared to negative controls. (TIF)

Porcine major histocompatibility complex (MHC) class I molecules and analysis of their peptide-binding specificities

In all vertebrate animals, CD8(+) cytotoxic T lymphocytes (CTLs) are controlled by major histocompatibility complex class I (MHC-I) molecules. These are highly polymorphic peptide receptors selecting and presenting endogenously derived epitopes to circulating CTLs. The polymorphism of the MHC effectively individualizes the immune response of each member of the species. We have recently developed efficient methods to generate recombinant human MHC-I (also known as human leukocyte antigen class I, HLA-I) molecules, accompanying peptide-binding assays and predictors, and HLA tetramers for specific CTL staining and manipulation. This has enabled a complete mapping of all HLA-I specificities (“the Human MHC Project”). Here, we demonstrate that these approaches can be applied to other species. We systematically transferred domains of the frequently expressed swine MHC-I molecule, SLA-1*0401, onto a HLA-I molecule (HLA-A*11:01), thereby generating recombinant human/swine chimeric MHC-I molecules as well as the intact SLA-1*0401 molecule. Biochemical peptide-binding assays and positional scanning combinatorial peptide libraries were used to analyze the peptide-binding motifs of these molecules. A pan-specific predictor of peptide–MHC-I binding, NetMHCpan, which was originally developed to cover the binding specificities of all known HLA-I molecules, was successfully used to predict the specificities of the SLA-1*0401 molecule as well as the porcine/human chimeric MHC-I molecules. These data indicate that it is possible to extend the biochemical and bioinformatics tools of the Human MHC Project to other vertebrate species. ELECTRONIC SUPPLEMENTARY MATERIAL: The online version of this article (doi:10.1007/s00251-011-0555-3) contains supplementary material, which is available to authorized users.

Major histocompatibility complex class I (MHC-I) molecules are found in all vertebrate animals where they play a crucial role in generating specific cellular immune responses against viruses and other intracellular pathogens. They are highly polymorphic proteins that bind 8-11 amino acid long peptides derived from the intracellular protein metabolism. The resulting heterotrimeric complexes-consisting of the MHC-I heavy chain, the monomorphic light chain, beta-2 microglobulin (β 2 m), and specifically bound peptides-are translocated to the cell surface where they displayed as target structures for peptide-specific, MHC-Irestricted CTLs. If a peptide of foreign origin is detected, the T cells may become activated and kill the infected target cell. MHC-I is extremely polymorphic. In humans, more than 3,400 different human leukocyte antigen class I (HLA-I) molecules have been registered (as of January 2011), and this number is currently growing rapidly as more efficient HLA typing techniques are employed worldwide. The polymorphism of the MHC-I molecule is concentrated in and around the peptide-binding groove, where it determines the peptide-binding specificity. Due to this polymorphism, it is highly unlikely that any two individuals will share the same set of HLA-I molecules thereby presenting the same peptides and generating T cell responses of the same specificities-something, that otherwise would give microorganisms a strong evolutionary chance of escape. Rather, this polymorphism can be seen as diversifying peptide presentation thereby individualizing T cell responses and reducing the risk that escape variants of microorganisms might evolve. In 1999, we proposed that all human MHC specificities should be mapped ("the Human MHC Project") as a preamble for the application of MHC information and technologies in humans (Buus 1999) . Since then, we have developed large-scale tools that are generally applicable towards this goal: production, analysis, prediction and validation of peptide-MHC interactions (Ferre et al. 2003; Harndahl et al. 2009; Hoof et al. 2009; Larsen et al. 2005; Lundegaard et al. 2008; Nielsen et al. 2003 Nielsen et al. , 2007 Ostergaard et al. 2001; Pedersen et al. 1995; Stranzl et al. 2010; Stryhn et al. 1996) , and a "one-pot, read-and-mix" HLA-I tetramer technology for specific T cell analysis (Leisner et al. 2008) . Here, we demonstrate that many of these tools can be transferred to other vertebrate animals as exemplified by an important livestock animal, the pig. We have successfully generated a recombinant swine leukocyte antigen I (SLA-I) protein, SLA-1*0401, one of the most common SLA molecules of swine (Smith et al. 2005) . Using this protein, we have developed the accompanying biochemical peptide-binding assays and demonstrated that the immunoinformatics tools originally developed to cover all HLA-I molecules, despite the evolutionary distance, can be applied to SLA-I molecules. We suggest that the "human MHC project" can be extended to cover other species of interest. All peptides were purchased from Schafer-N, Denmark (www.schafer-n.com). Briefly, they were synthesized by standard 9-fluorenylmethyloxycarbonyl (Fmoc) chemistry, purified by reversed-phase high-performance liquid chromatography (to at least >80% purity, frequently 95-99% purity), validated by mass spectrometry, and quantitated by weight. Positional scanning combinatorial peptide libraries (PSCPL) peptides were synthesized using standard solidphase Fmoc chemistry on 2-chlorotrityl chloride resins. Briefly, an equimolar mixture of 19 of the common Fmoc amino acids (excluding cysteine) was prepared for each synthesis and used for coupling in 8 positions, whereas a single type of Fmoc amino acid (including cysteine) was used in one position. This position was changed in each synthesis starting with the N-terminus and ending with the C-terminus. In one synthesis, the amino acid pool was used in all nine positions. X denotes the random incorporation of amino acids from the mixture, whereas the single letter amino acid abbreviation is used to denote identity of the fixed amino acid. The peptides in each synthesis were cleaved from the resin in trifluoroacetic acid/1,2-ethanedithiol/triisopropylsilane/water 95:2:1:3 v/v/v/v, precipitated in cold diethylether, and extracted with water before desalting on C18 columns, freeze drying, and weighting. Recombinant constructs encoding chimeric and SLA-1*0401 molecules A synthetic gene encoding a transmembrane-truncated fragment encompassing residues 1 to 275 of human HLA-A*11:01 alpha chain followed by a FXa-BSP-HAT tag (FXa = factor Xa cleavage site comprised of the amino acid sequence IEGR, BSP = biotinylation signal peptide, HAT = histidine affinity tag for purification purposes; see Online Resource 1) had previously been generated and inserted into the pET28 expression plasmid (Novagen) (Ferre et al. 2003) . Synthetic genes encoding the corresponding fragments of the SLA-1*0401 alpha chain (α 1 α 2 ) and α 3 , respectively, (Sullivan et al. 1997) were purchased from GenScript. To exchange domains and generate chimeric human/swine MHC-I gene constructs, a type II restriction endonuclease-based cloning strategy (SeamLess® Strategene; Cat#214400, Revision#021003a), with modifications, was used. All primers were purchased HPLC-purified from Eurofins MWG Operon (Ebersberg, Germany), and all PCR amplifications were performed in a DNA Engine Dyad PCR instrument (MJ Research, MN, USA). All constructs were validated by DNA sequencing. The following MHC-I heavy chain constructs were made HHH, HHP, HPP, PHP, and PPP, where the first, second, and third letter indicates domains α 1 (positions 1-90), α 2 (positions 91-181), and α 3 (positions 182-275), respectively, and H indicates that the domain is of HLA-A*11:01 origin, whereas P indicates that it is of SLA-1*0401 origin. Constructs were transformed into DH5α cells, cloned, and sequenced (ABI Prism 3100Avant, Applied Biosystems) . Validated constructs of interest were transformed into an Escherichia coli production cell line, BL21(DE3), containing the pACYC184 expression plasmid (Avidity, Denver, USA) containing an isopropylβ-d-1-thiogalactopyranoside (IPTG)-inducible BirA gene to express biotin-ligase. This leads to almost complete in vivo biotinylation of the desired product (Leisner et al. 2008 ). To maintain the pET28-derived plasmids, the media was supplemented with kanamycin (50 μg/ml) throughout the expression cultures. When appropriate, the media was further supplemented with chloroamphenicol (20 μg/ml) to maintain the BirA containing pACYC184 plasmid. E. coli BL21(DE3) cells transformed with appropriate plasmids were grown for 5 h at 30°C, and a 10-ml sample adjusted to OD (600) =1 was then transferred to a 2-l fedbatch fermentor (LabFors®). To induce protein expression, IPTG (1 mM) was added at OD (600) ∼25 and the culture was continued for an additional 3 h at 42°C (for in vivo biotinylation of the product, the induction media was further supplemented with biotin (Sigma #B4501, 125 μg/ ml)). Samples were analyzed by reducing sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) before and after IPTG induction. At the end of the induction culture, protease inhibitor (PMSF, 80 μg/l) was added, and cells were lysed in a cell disrupter (Constant Cell Disruptor Systems set at 2,300 bar) and the released inclusion bodies were isolated by centrifugation (Sorval RC6, 20 min, 17,000×g). The inclusion bodies were washed twice in PBS, 0.5% NP-40 (Sigma), and 0.1% deoxycholic acid (Sigma) and extracted into urea-Tris buffer (8 M urea, 25 mM Tris, pH 8.0), and any contaminating DNA was precipitated with streptomycin sulfate (1% w/v). The dissolved MHC-I proteins were purified by Ni 2 +/IDA metal chelating affinity column chromatography followed by Q-Sepharose ion exchange column chromatography, hydrophobic interaction chromatography, and eventually by Superdex-200 size exclusion chromatography. Fractions containing MHC-I heavy chain molecules were identified by A280 absorbance and SDS-PAGE and pooled. Throughout purification and storage, the MHC-I heavy chain proteins were dissolved in 8 M urea to keep them denatured. Note that the MHC-I heavy chain proteins at no time were exposed to reducing conditions. This allowed purification of highly active pre-oxidized moieties as previously described (Ostergaard et al. 2001) . Protein concentrations were determined by bicinchoninic acid assay. The degree of biotinylation (usually >95%) was determined by a gel-shift assay (Leisner et al. 2008 ). The pre-oxidized, denatured proteins were stored at −20°C in Tris-buffered 8 M urea. Recombinant constructs encoding human and porcine beta-2 microglobulin Recombinant human β 2 m was expressed and purified as described elsewhere (Ostergaard et al. 2001) , (Ferre et al. 2003) . Using a previously reported E. coli codonoptimized gene encoding human β 2 m as template , a gene encoding porcine β 2 m was generated by multiple rounds of site-directed mutagenesis (QuikChange® Stratagene, according to the manufacturer's instructions) (Online Resource 2). Briefly, the genes encoding human or pig β 2 m were N-terminally fused to a histidine affinity encoding tag (HAT) followed by a restriction enzyme encoding tag (FXa), inserted into the pET28 vector and expressed in inclusion bodies in E. coli. The fusion proteins were extracted into 8 M urea, purified by immobilized metal affinity chromatography (IMAC), and refolded by dilution. The fusion tags were then removed by FXa restriction protease digestion. The liberated intact and native human or pig β 2 m were purified by IMAC and gel filtration chromatography, analyzed by SDS-PAGE analysis, concentrated, and stored at −20°C until use (Fig. 1 ). Purification and refolding of recombinant porcine β 2 m proteins Porcine β 2 m was purified in the same way as human β 2 m (Ostergaard et al. 2001; Ferre et al. 2003) . Briefly, the ureadissolved β 2 m protein was purified by Ni 2 +/IDA metal chelating affinity column chromatography, refolded by drop-wise dilution into an excess refolding buffer under stirring (25 mM Tris, 300 mM urea, pH 8.00), and then concentrated (VivaFlow, 10 kDa). The refolded product was purified by Ni 2 +/IDA metal chelating affinity column chromatography again (this time in aqueous buffer, i.e., without urea). Fractions containing HAT-pβ 2 m were iden-tified by SDS-PAGE and pooled. Removal of the HAT tag was performed by cleavage with factor Xa restriction protease (FXa) followed by renewed purified by Ni 2 +/ IDA metal chelating affinity and Superdex200 gel filtration column chromatography, concentrated by spin ultrafiltration (10 kDa), mixed 1:1 with glycerol, and stored at −20°C. Protein samples were mixed 1:1 in SDS sample buffer (4% SDS, 17.4% glycerol, 0.003% bromophenol blue, 0.125 M Tris, 8 mM IAA (iodoacetamide)) with or without reducing agent (2-mercaptoethanol) as indicated, boiled for 3 min, spun at 20,000×g for 1 min, and loaded onto a 12% or 15% running gel with a 5% stacking gel. Gels were run at 180 V, 40 mA for 50 min. Peptide-MHC class I interaction measured by radioassay and spun column chromatography A HLA-A*11:01-binding peptide, KVFPYALINK (nonnatural consensus sequence A3CON1 ), was radiolabeled with iodine ( 125 I) using a chloramine-T procedure (Hunter and Greenwood 1962) . Dose titrations of MHC-I heavy chains (HHH or HHP) were diluted into refolding buffer (Tris-maleate-PBS) and mixed with β 2 m (human or porcine) and radiolabeled peptide, and incubated at 18°C overnight. Then binding of radiolabeled peptide to MHC-I was determined in duplicate by Sephadex™ G50 spun column gel chromatography as previously described . MHC bound peptide eluted in the excluded volume, whereas free peptide was retained on the microcolumn. Both fractions were counted by gamma spectroscopy, and the fraction peptide bound was calculated as excluded radioactivity divided by total radioactivity. To examine the affinity of the interaction, increasing concentrations of unlabeled competitor peptide were added. When conducted under limiting concentrations of MHC-I molecule, the concentration of competitor peptide needed to effect 50% inhibition of the interaction, the IC 50 , is an approximation of the affinity of the interaction between MHC-I and the competitor peptide. Peptide-MHC class I interaction measured by an enzyme-linked immunosorbent assay Peptide-MHC-I interaction was also measured in a modified version of a previously described enzyme-linked immunosorbent assay (ELISA) (Sylvester-Hvid et al. 2002) . Briefly, denatured biotinylated recombinant MHC-I heavy chains were diluted into a renaturation buffer containing β 2 m and graded concentrations of the peptide to be tested and incubated at 18°C for 48 h allowing equilibrium to be reached. We have previously demonstrated that denatured MHC molecules can de novo fold efficiently, however, only in the presence of appropriate peptide. The concentration of peptide-MHC complexes generated was measured in a quantitative sandwich ELISA (using streptavidin as capture layer and the monoclonal anti-β 2 m antibody, BBM1, as detection layer) and plotted against the concentration of peptide offered (Sylvester-Hvid et al. 2002) . A prefolded, biotinylated FLPSDYFPSV/ HLA-A*02:01 (Kast et al. 1994 ) complex was used as standard. Because the effective concentration of MHC (3-5 nM) used in these assays is below the equilibrium dissociation constant (K D ) of most high-affinity peptide-MHC interactions, the peptide concentration, ED 50 , leading to half-saturation of the MHC is a reasonable approximation of the affinity of the interaction. The experimental strategy of PSCPL has previously been described (Stryhn et al. 1996) . The construction of the sublibraries and the ELISA-driven quantitative measurements of MHC interaction are as given above. Briefly, the relative binding (RB) affinity of each PSCPL sublibrary was determined as RB (PSCPL)=ED 50 (X 9 )/ ED 50 (PSCPL) (where ED 50 is the concentration needed to half-saturate a low concentration of MHC-I molecules) and normalized so that the sum of the RB values of the 20 naturally occurring amino acids equals 20 (since peptides with a given amino acid in a given position are 20 times more frequent in the corresponding PSCPL sublibrary than in the completely random X 9 library). A RB value above 2 was considered as the corresponding position and amino acid being favored, whereas a RB value below 0.5 was considered as being unfavorable (these thresholds represent the 95% confidence intervals). An anchor position (AP) value was calculated by the equation ∑(RB−1) 2 . A primary anchor position is characterized by one or few amino acids being strongly preferred and many amino acids being unacceptable. We have arbitrarily defined anchor residues as having an AP value above 15 ). The peptide-SLA-I*0401 binding activity of each sublibrary was determined using previously published biochemical binding assay (ELISA) (Sylvester-Hvid et al. 2002 ) (with the modifications described above). Sequences logos describing the predicted binding motif for each MHC molecule were calculated as described by Rapin et al. (2010) . In short, the binding affinity for a set of 1,000,000 random natural 9mer peptides was predicted using the NetMHCpan method, and the 1% strongest binding peptides were selected for construction of a position-specific scoring matrix (PSSM). The PSSM was constructed as previously described including pseudo-count correction for low counts. Next, sequence logos were generated from the amino acid frequencies identified in the PSSM construction. For each position, the frequency of all 20 amino acids is displayed as a stack of letters. The total height of the stack represents the sequence conservation (the information content), while the individual height of the symbols relates to the relative frequency of that particular symbol at that position. Letter shown upside-down are underrepresented compared to the background (for details see Rapin et al. (2010) ). MHC distance trees were derived from correlations between predicted binding affinities. For each allelic MHC-I molecule, the binding affinity was predicted for a set of 200,000 random natural peptides using the NetMHCpan method. Next, the distance between any two alleles was defined, as D = 1− PCC, where PCC is the Pearson correlation between the subset of peptides within the superset of top 10% best binding peptides for each allele. In this measure, two molecules that share a similar binding specificity will have a distance close to 0 whereas two molecules with non-overlapping binding specificities would have a distance close to 2. Using bootstrap, 100 such distance trees were generated, and branch bootstrap values and the consensus tree were calculated. We have previously generated highly active, recombinant human MHC-I (HLA-I) molecules and accompanying highthroughput assays and bioinformatics prediction resources. Here, we transfer the underlying approaches to an important domesticated livestock animal, the pig, and its MHC system, the SLAs. MHC-I molecules are composed of a unique and highly variable distal peptide-binding platform consisting of the alpha 1 (α 1 ) and alpha 2 (α 2 ) domains of the MHC-I heavy chain (HC) and a much more conserved proximal immunoglobulin-like membrane attaching stalk consisting of the alpha 3 (α 3 ) domain of the HC noncovalently associated with the soluble MHC-I light chain (β 2 m). A priori, the establishment of recombinant SLA molecules is complicated by the lack of validated reagents. Any failure could therefore be caused either by real technical problems in generating SLA molecules, or merely by a lack of information about strong peptide binders to the SLA in question. To reduce this uncertainty, we decided to migrate from human to pig MHC-I in a step-wise manner and generate an intermediary chimeric MHC-I molecule composed of a well-known human peptide-binding platform attached to a SLA stalk, which might allow us to assess whether we could generate a functional SLA stalk consisting of SLA-1*0401 α 3 HC and pig β 2 m. To this end, we used the α 1 α 2 domains of the HLA-A*11:01 molecule, which we expected should be able to bind a known highaffinity HLA-A*11:01-binding peptide (KVFPYALINK). This peptide could be 125 I radiolabeled and used in a very robust peptide-binding assay testing whether the human stalk could be replaced with the corresponding SLA stalk. Once that had been successfully established, the entire SLA-1*0401 molecule would be constructed and tested. We have previously expressed and purified the extracellular segment spanning positions 1-275 of the human HLA-A*11:01 in a denatured and pre-oxidized version that rapidly refold and bind appropriate target peptides (Ostergaard et al. 2001; Ferre et al. 2003) . Codon-optimized genes encoding the corresponding segments of SLA-1*0401 (α 1 α 2 ) and SLA-1*0401 (α 3 ) were constructed as described in the "Materials and methods" section and used to replace the HLA-A*11:01 gene segment in the above construct generating a new construct allowing for the expression of SLA-1*0401. For the generation of HLA-A*11:01/SLA-1*0401 chimeras, the genes encoding α 1 (spanning positions 1-90), α 2 (spanning positions 91-181), and/or α 3 (spanning positions 182-275) domains of HLA-A*11:01 and SLA-1*0401 were exchanged using Seam-Less and touch-down cloning strategies. Genes encoding the extracellular segments 1-275 of the above natural or chimeric MHC-I molecules were C-terminally fused to a biotinylation tag (as indicated for SLA-1*0401 in Online Resource 1), inserted into pET28, and expressed in inclusion bodies in E. coli (Fig. 2 shows SDS-PAGE of lysates of recombinant E. coli before and 3 h after IPTG induction). The fusion proteins were extracted into 8 M urea (without any reducing agents), purified by ion exchange, hydrophobic and gel filtration chromatography (all conducted in 8 M urea, without any reducing agents) (Fig. 3 shows SDS-PAGE of the purified SLA-1*0401 after gel filtration), concentrated, and stored in urea at −20°C. Testing a chimeric molecule consisting of a SLA-1*0401 stalk and a HLA-A*11:01 peptide-binding platform-comparing human versus porcine β 2 m To test the proximal immunoglobulin-like membrane attaching SLA stalk, we generated recombinant porcine β 2 m and a chimeric human/porcine MHC-I heavy chain molecule where the α 1 α 2 were derived from the human HLA-A*11:01, and the α 3 was derived from the porcine SLA-1*0401. Since this construct contains the entire peptide-binding platform of HLA-A*11:01, we reasoned that the binding of the HLA-A*11:01 restricted peptide, KVFPYALINK, could be used as a functional readout of the refolding, activity, and assembly of the entire chimeric molecule including the porcine SLA stalk. For comparison, we tested the supportive capacity of human β 2 m and folding ability of the entirely human HLA-A*11:01. A total of four combinations could therefore be tested: porcine or human β 2 m in combination with either HHP or HHH (where the first letter indicates the origin of the α 1 domain (Human HLA-A*11:01 or Porcine SLA-1*0401), the second letter the origin of the α 2 domain, and the third letter the origin of the α 3 domain). A concentrationtitration of heavy chain was added to a fixed excess concentration (3 μM) of β 2 m and a fixed trace concentration (23 nM) of radiolabeled peptide. As shown in Figs. 4 and 5, the four combinations gave almost the same heavy chain dose titration with a half-saturation occurring around 1-2 nM heavy chain. Porcine β 2 m supported folding of the chimeric (HHP) α chain slightly better than it supported folding of the human (HHH) α chain. Human β 2 m supported folding of HHP and HHH equally well. Thus, a recombinant SLA stalk can fold and support peptide binding of the peptide-binding platform. These results also suggest that human β 2 m can support folding and peptide binding of porcine MHC-I heavy chain molecules. Using a positional scanning combinatorial peptide library approach to perform an unbiased analysis of the specificity of SLA-1*0401 and human-pig chimeric MHC class I molecules Using human β 2 m to support folding, the recombinant SLA-1*0401 and human-pig chimeric MHC-I molecule were tested for peptide binding. We have previously described how PSCPL can be used to perform an unbiased analysis of MHC-I molecules (Stryhn et al. 1996) . A PSCPL consists of 20 sublibraries for each position where one of each of the 20 natural amino acids have been locked and all other positions contain random amino acids. Analyzing how much of each PSCPL sublibrary is needed to support MHC-I folding (see examples in Fig. 6 ) and comparing each sublibrary with a completely random library, the effect of any amino acid in any position can be examined and expressed as a RB value. Further, an AP value calculated as the sum of squared deviations of RB values for each position can be used to identify the most prominent anchor position (see "Materials and methods" for calculations). Thus, the specificity of a nonamer binding MHC-I molecule can be analyzed comprehensively with 9×20+1 completely random library=181 sublibraries (Stryhn et al. 1996) . Here, this approach was used to perform a complete experimental analysis of SLA-1*0401 and a limited analysis of the chimeric HPP and PHP molecules. A nonamer PSCPL analysis of SLA-1*0401 can be seen in Table 1 . AP values identified positions 9, 3, and 2 (in that order of importance) as the anchor positions of SLA-1*0401. In position 9, the amino acid preferences were dominated by the large and bulky aromatic tyrosine (Y), tryptophane (W), and phenylalanine (F), all having RB values above 4 (Table 1 ). In the almost equally important position 3, preferences for negatively charged amino acids glutamic acid (E) and aspartic acid (D) were observed. In the lesser important position 2, the most preferred amino acids were the hydrophobic amino acids valine (V), isoleucine (I), and leucine (L), followed by the polar amino acids threonine (T) and serine (S). Finally, a very limited PSCPL analysis was performed for the two chimeric human HLA-A*11:01/porcine SLA-1*0401 MHC-I molecules, HPP and PHP (Table 2) . For both chimeric molecules, it could be demonstrated that position 9 is a strong anchor position. The positively charged amino acids, arginine (R) and lysine (K), were preferred in position 9 of the chimeric HPP molecule, whereas the aromatic amino acid, tyrosine (Y), was exclusively preferred in position 9 of the chimeric PHP molecule. The positively charged amino acids, arginine (R) and lysine (K), were preferred in position 9 of the chimeric HPP molecule similar to the position 9 specificity of the HLA-A*11:01 molecule. In contrast, the aromatic amino acid tyrosine (Y) was preferred in position 9 of the chimeric PHP molecule similar to the position 9 specificity of the SLA-1*0401 molecule. Using NetMHCpan to predict peptides that bind to SLA-1*0401 or to human-pig chimeric MHC class I molecules Our recently described neural network-driven bioinformatics predictor, NetMHCpan (version 2.0), has been trained on about 88,000 peptide-binding data points representing more than 80 different MHC-I molecules (primarily HLA-A and HLA-B molecules). We have previously shown that NetMHCpan is an efficient tool to identify peptides that bind to HLA molecules where no prior data exist (Nielsen et al. 2007 ) and demonstrated that NetMHCpan can be extended to MHC-I molecules of other species 1 (Hoof et al. 2009 ). We applied NetMHCpan to our peptide repository of about 10,000 peptides, which over the past decade have been selected to scan infectious agents (e.g., SARS and influenza, Sylvester-Hvid et al. 2004; Wang et al. 2010) , improve coverage of MHC-I specificities (e.g., Buus et al. 2003; Christensen et al. 2003) , etc. We extracted 29 peptides as predicted binders to either the SLA-1*0401, the HPP, or the PHP human/porcine chimeric class I molecules (some of the peptides were predicted to bind to two or even all three of these molecules). All these peptide-MHC-I combinations were tested for binding (Table 3) ; 13 of 14 peptides bound to the SLA-1*0401 molecule with an affinity (IC 50 value) better than 500 nM (6 with an affinity less than 50 nM); all 13 peptides tested on the PHP molecule were strong binders with IC 50 values below 50 nM; and 3 of 12 peptides tested on the HPP molecule bound with an affinity better than 500 nM. Of the 39 peptide-MHC-I combinations tested, 20 (51%) were found to be good binders, 9 (23%) were average binders, and 10 (26%) did not bind well (Table 3) . This is in stark contrast to the 0.5% frequency of binders among randomly selected peptides (Yewdell and Bennink 1999) . Next, the NetMHCpan method was used to generate PSSMs and sequence logos from the corresponding amino acid frequencies as described by Nielsen et al. (2004) . For each position, the frequencies of all 20 amino acids were displayed as a stack of letters showing the sequence conservation/information content (the height of the entire 1 A preliminary report of SLA-1*0401 binding was given in Hoof et al. (2009) The normalized relative binding (RB) value indicates whether an amino acid is favored (RB>2, bold numbers) or disfavored (RB<0.5, italic numbers) in a given peptide position. The anchor position (AP) value is given by the equation ∑(RB−1) 2 . The important anchor positions 2, 3, and 9 for SLA-1*0401 are underlined stack) and the relative frequency of amino acids (the height of the individual amino acids). Figure 7 shows a specificity tree clustering of the SLA-1*0401 molecule compared to prevalent representatives of the 12 common HLA supertypes that NetMHCpan originally intended to cover . By this token, SLA-1*0401 most closely resembles that of HLA-A*01:01. The limited PSCPL analysis of the chimeric MHC-I molecules revealed strong P9 signals with specificities that seemed to be determined by the origin of the α 1 domain: the HPP chimera showed an HLA-A*11:01-like P9 specificity, whereas the PHP chimera showed a SLA-1*0401/HLA-A*01:01-like specificity. Since the NetMHCpan predictor successfully captured these chimeric specificities (see above), we reasoned that the predictor might also be used to perform in silico dissection of these specificities and used the P9 specificity as an example of such an in silico analysis. The NetMHCpan predictor considers a pseudo-sequence consisting of 34 polymorphic positions, which contain residues that are within 4.0 Å of the atoms of bound nonamer peptides (Nielsen et al. 2007 ). Of the 34 positions of the pseudosequence, 10 delineates the P9 binding pocket; however, only 3 of these, positions 74, 77, and 97, differ between SLA-1*0401 and HLA-A*11:01. To explore the effect of these three residues, we performed in silico experiments where we examined single substitutions Y74D, G77D, and S97I (the letter before the position number indicates the SLA-A*0401 single letter residue, whereas the letter after indicates the HLA-A*11:01 residue) as well as the corresponding triple substitution (YGS-DDI). As described above, PSSMs were generated for each of the in silico molecules followed by a specificity tree clustering (including SLA-A*0401, HLA-A*01:01, and HLA-A*11:01). Figure 8 shows this tree along with the sequence logo plots showing the predicted binding specificity of each in silico MHC-I molecule. Albeit the Y74D and G77D single substitutions showing some of the positively charged P9 peptide residue preference of HLA-A*11:01, they still clustered with HLA-A*01:01. In contrast, the in silico (YGS-DDI) triple substitution clustered with the HLA-A*11:01. This suggests that the NetMHCpan method is capable of defining the residues of the F pocket that determine the specificity of position 9. We have previously suggested that the specificities of the entire human MHC-I system should be solved ("the human RB and AP values are defined as described in Table 1 MHC", Buus 1999; Lauemoller et al. 2000) . However, due to the extreme polymorphism of the MHCs, any attempt to address the specificity of the entire MHC system is a significant experimental undertaking. During the past decade, we have established a series of technologies to support a general solution of human MHC class I and II specificities. For MHC-I, this includes (1) a highly efficient E. coli expression system for production of recombinant human and mouse MHC-I molecules (both heavy chain and light chain (β 2 m) molecules) Ostergaard et al. 2001) , (2) a purification system for obtaining the highly active pre-oxidized MHC-I heavy chain species (Ferre et al. 2003) , (3) a high-throughput homogenous peptide-MHC-I binding assay for obtaining large data sets on peptide-MHC-I interactions (Harndahl et al. 2009 ), (4) a positional scanning combinatorial peptide library approach for a robust and unbiased analysis of the specificity of any MHC-I molecule (Stryhn et al. 1996) , (5) an immunobioinformatics approach to generate predictors of the peptide-MHC-I interaction, NetMHCpan, that allows predictions to be made for any human MHC-I molecules, HLA-I, even those that have not yet been covered by existing data set (Hoof et al. 2009; Nielsen et al. 2007) , and finally (6) we have demonstrated that pre-oxidized MHC-I molecules can be used to generate MHC-I tetramers in a simple "one-pot, mix-and-read" manner (Leisner et al. 2008) . Here, we propose that the next goal should be to extend the overall approach to MHC-I molecules of other species of interest. Mouse and rats have been extensively studied in the past, but much less reagents and information have accrued for the MHC-I molecules of other species. Here, we have used an important livestock animal, the pig, as a model system and demonstrated that it indeed is possible to transfer the original human approach to other species. Before attempting to generate a recombinant version of the entire porcine SLA-1*0401 molecule, we grafted the more conserved membrane-proximal "stalk" (the immunoglobulin-like class I heavy chain α 3 and β 2 m domains) of porcine SLA-1*0401 onto the peptide-binding domain of HLA-A*11:01 generating a chimeric human/ porcine MHC-I molecule. This chimeric molecule retained the peptide-binding specificity of the HLA-A*11:01 molecule, and it clearly demonstrated that the recombinant porcine stalk was functional and, by inference, also properly folded. It also suggests that the peptide-binding specificity of the distal domains do not crucially depend upon the identity of the proximal stalk. Further, comparing the ability of human and porcine β 2 m to support MHC-I complex formation using either a human or a porcine MHC-I stalk, we demonstrated that every combination (porcine β 2 m/human-α 3 , porcine β 2 m/porcine-α 3 , human β 2 m/human-α 3 , and human β 2 m/porcine-α 3 ) showed al- most the same heavy chain dose titration with identical half-saturations. These results illustrate the ability for porcine and human β 2 m to support complex formation of SLA molecules and vice versa and suggest evolutionary that the stalk is quite conserved. Next, we generated the entire SLA-1*0401 heavy chain and succeeded in generating complexes using human β 2 m as the light chain and PCSPL as peptide donors. The latter solved the a priori problem of not knowing which peptides would be needed to support proper folding of SLA-1*0401, and it did so in an unbiased manner. Furthermore, this approach is highly efficient since it readily establishes a complete matrix representing the amino acid preference for each amino acid and each position of a nonamer peptide. The specificity of SLA-1*0401 shows two primary anchors: one in positions 9 with a preference for aromatic amino acids and another in position 3 with a preference for negatively charged amino acids. In addition, the SLA-1*0401 features a secondary anchor in position 2 with hydrophobic or polar amino acid preferences. An alternative approach to solve the problem of identifying peptides that support folding of MHC-I molecules of so far unknown specificity is to use our recently developed panspecific predictor, NetMHCpan. The successful use of this predictor to initiate peptide-binding studies was recently Fig. 7 Specificity tree clustering of the SLA-1*0401 molecule compared to prevalent representatives of the 12 common HLA supertypes ). The distance between any two MHC molecules and the consensus tree is calculated as described in "Materials and methods". All branch points in the tree have bootstrap values of 100%. Sequence logos of the predicted binding specificity are shown for each molecule. In the logo, acidic amino acids [DE] are shown in red, basic amino acids [HKR] in blue, hydrophobic amino acids [ACFILMPVW] in black, and neutral amino acids [GNQSTY] in green. The axis of the LOGOs indicates in all case positions one through nine of the motif, and the y-axis the information content (see Materials and methods) demonstrated for HLA-A*3001 . Although originally developed to cover all HLA-A and HLA-B molecules, it has also been shown to extend to MHC-I molecules of other species (Hoof et al. 2009 ). Here, we demonstrate that the NetMHCpan predictor is capable of extracting MHC-I sequence information across species and correctly relate this to peptide binding even in the absence of any available data for the specific query MHC-I molecule, i.e., the SLA-1*0401 as well as the chimeric HPP (hα 1 pα 2 pα 3 ) and PHP (pα 1 hα 2 pα 3 ) molecules. It is not clear why binding of the PHP chimera was more efficiently predicted than binding of the HPP chimera. One could speculate that NetMHCpan has not captured the effect of the different positions of the pseudo-sequence equally well and not all positions and pockets (and by inference-not all chimeric molecules) are therefore predicted equally well. Using the NetMHCpan predictor to cluster SLA-1*0401 and representative molecules of 12 human HLA supertypes according to predicted peptide-binding specificities, the SLA-1*0401 specificity closely resembled that of HLA-A*01:01 (IEDB, http://www.immuneepitope.org/MHCalleleId/142, accessed March 9th 2011). This result was also obvious from an inspection of the PSCPL analysis of the SLA-1*0401. The PSCPL analysis of the P9 specificity of the SLA-1*0401 and the two chimeric molecules suggested that the P9 specificity primarily was determined by the α 1 domain. This contention was further strengthened by a NetMHCpan-driven in silico analysis of the residues delineating the F pocket, which interacts with P9. This suggests that NetMHCpan can be used to design and interpret detailed experiments addressing the structurefunction relationship of peptide-MHC-I interaction. In the case of SLA-1*0401, NetMHCpan suggests that Y74, G77, and S97 play a prominent role in defining the P9 F pocket. Whereas the NetMHCpan readily captured the P9 anchor residue of SLA-1*0401, it did not capture the P3 anchor (at least not in the 2.4 version). We surmise that this shortcoming is due to insufficient examples of the use of P3 anchors within the currently available peptide-MHC-I binding data. Inspecting the pseudo-sequence of SLA-1*0401 and HLA-A*01:01 vs. HLA-A*11:01 suggests that the presence of an arginine in position 156 might Fig. 8 Comparison of specific in silico mutations of the SLA-1*0401 molecule and comparison with the two HLA molecules: HLA-A*11:01 and HLA-A*01:01. The distance between any two MHC molecules and the consensus tree is calculated as described in "Materials and methods". All branch points in the tree have bootstrap values of 100%. The SLA-1*0401 mutations are indicated as Y74D, G77D, and S97I, where the letter before the position number indicates the SLA-1*0401 single letter residue and the letter after indicates the HLA-A*11:01 residue. YGS-DDI is the corresponding triple substitution. Sequence logos are calculated and visualized as described in Fig. 7 . The axis of the LOGOs indicates in all case positions one through nine of the motif, and the y-axis the information content (see Materials and methods) explain the preference for negatively charged amino acid residues in P3. Future NetMHCpan-guided experiments could pointedly address this question, and the resulting data could complement existing data and be used to update and improve the NetMHCpan predictor. All in all the two complementary approaches, PSCPL and NetMHCpan, agreed on the specificity of the SLA-1*0401 molecule, as well as of the two chimeric MHC-I molecules. Thus, the specificity of SLA-1*0401 appear to be well established. This specificity has successfully been used to search for foot-and-mouth disease virus (FMDV)specific CTL epitopes in FMDV-vaccinated, SLA-1*0401positive pigs, and the recombinant SLA-1*0401 molecules have been used to generate corresponding tetramers and stain pig CTLs (Patch et al. 2011) . In conclusion, we here present a set of methods that can be used to generate functional recombinant MHC-I molecules, map their specificities and identify MHC-I-restricted epitopes, and eventually generate peptide-MHC-I tetramers for validation of CTL responses. This suite of methods is not only applicable to humans, but potentially to any species of interest.

Immunogenetic Factors Associated with Severe Respiratory Illness Caused by Zoonotic H1N1 and H5N1 Influenza Viruses

Following the 2009 H1N1 pandemic and ongoing sporadic avian-to-human transmission of H5N1 viruses, an emphasis has been placed on better understanding the determinants and pathogenesis of severe influenza infections. Much of the current literature has focused on viral genetics and its impact on host immunity as well as novel risk factors for severe infection (particularly within the H1N1 pandemic). An understanding of the host genetic determinants of susceptibility and severe respiratory illness, however, is currently lacking. By better defining the role of genetic variability in influenza infection and identifying key polymorphisms that impair the host immune response or correlate with protection, we will be able to better identify at-risk populations and new targets for therapeutic interventions and vaccines. This paper will summarize known immunogenetic factors associated with susceptibility or severity of both pH1N1 and H5N1 infections and will also identify genetic pathways and polymorphisms of high relevance for future study.

Transmission of zoonotic influenza A viruses to humans is commonly the cause of new pandemics, which typically result in high disease burden and increased symptomatic severity and mortality. In order to predict which populations may be at highest risk of infection and to develop more effective therapeutic interventions and vaccines, a thorough understanding of both viral and host contribution to pathogenesis is required. In both the recent 2009 H1N1 (pH1N1) pandemic and the on-going rare avian-to-human transmission of H5N1, numerous studies have taken an indepth look at the impact of viral evolution and mutation on viral pathogenesis. Conversely, while both human and animal model studies of the host immune response to infection have identified correlates of severe disease, the contribution of host genetics to these correlates and to variability in susceptibility remains relatively unknown. Identification of host genetic polymorphisms contributing to altered susceptibility or disease severity has several benefits: identification of high-risk populations at greater need of prophylactic intervention, elucidation of host proteins important in virus-host interactions, and new targets for therapeutic interventions or vaccine development [1] . Studies of host genetics have provided important contributions to the study of other infectious diseases, including HIV, SARS, and HCV. This paper will describe what is currently known about the impact of host immunogenetics in both pH1N1 and H5N1 infections and will identify highly relevant polymorphisms and genetic pathways that could be investigated in future work. H1N1 influenza viruses emerged as a result of a presumed or documented reassortment of segments from viruses of zoonotic origin with human-adapted influenza virus to cause pandemic spread in 1918 and again in 2009. The 2009 appearance of a swine-origin reassortant virus led to the first pandemic of the 21st century. During earlier pandemics, records indicate that certain individuals or populations appeared to be more susceptible to severe disease, but the 2 Clinical and Developmental Immunology ability to conduct studies in order to understand the immune mechanisms that underlay the increased propensity for complications was limited. The 2009 H1N1 (pH1N1) pandemic was accompanied by improved surveillance, thereby facilitating better estimation of disease severity and methods to examine the immune mechanisms behind complicated disease [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] . This surveillance allowed for the identification of several novel risk factors among various populations, but with a limited understanding of the genetic variation that may contribute to those risk factors. Infection. The 1918 H1N1 as well as the recent 2009 pandemics were both notable for the comparatively high rates of morbidity among healthy, young adults not typically observed with seasonal influenza [11] . During the recent pandemic, several studies of confirmed pH1N1 cases in Canada and the US reported the median age of severe infections to be 23-27 years old [8, 10] . In Canada, 30%-48% of infections also presented in persons with comorbidities; diabetes, heart disease, and immunosuppression were associated with the highest risk of severe infection, while lung diseases and obesity were among the most common underlying conditions [10, [12] [13] [14] . The role of pregnancy as a risk factor, regardless of the stage, was also supported by a myriad of reports; among hospital admissions, pregnancy accounted for roughly 30% of female cases aged 20-39 years old [9, 12, 15] . Ethnicity was another major risk factor of pH1N1 susceptibility identified in several populations in North America and Australasia. The increased proportion of aboriginal individuals presenting with severe pH1N1 infection was not unique for this pandemic and was also seen in the 1918 H1N1 pandemic during which mortality in aboriginal communities in North America (3%-9%) was significantly higher than among nonaboriginal communities [16, 17] . In the 2009 pandemic, Pacific Islanders accounted for 2.5% of the Australian population but made up 9.7% of patients admitted to Australian ICUs with confirmed pH1N1. Maori individuals represent 13.6% of the New Zealand population, but accounted for 25% of ICU admissions in the ANZIC study [18] . Kumar et al. [12] also reported 25.6% of the individuals admitted to ICUs in Canada belonged to First Nations, Inuit, Metis, or aboriginal ethnicities; this is an overrepresentation compared to the 4.4% rate of selfreported aboriginal ethnicity according to the 2001 census (Statistics Canada). Similarly, pH1N1 mortality rates among American-Indian/Alaska Natives were four times higher than persons in all other ethnic populations combined in the United States [19] . None of these studies examined the causal factors that lead to the higher influenza mortality in the high-risk groups described. It is clear that multiple converging risks account for the high rates of complications, including socioeconomic factors such as inability to access care, delayed seeking of care, higher rates of poverty, and greater numbers of household members. A few of the risk factors listed in the previous sections, however, share a degree of immune system impairment. One can, therefore, speculate that the partial protection afforded by the immune system, primarily by cross-reactive CD8+ T-cells recognizing viral epitopes, is decreased in some of the previously described groups (pregnancy is a good example). Additionally, genetic variation in immune-related genes leading to either gain-offunction or loss-of-function phenotypes could contribute to the variation observed in pH1N1 susceptibility and disease severity. Associated with Severity of Pandemic H1N1 Infection. When a novel strain of influenza emerges, the pre-existing antibody response directed largely at the surface glycoproteins is rendered ineffective. In these cases, the mechanisms underlying heterosubtypic cross-protection assume a dominant role and it is, therefore, not surprising that immune dysfunction caused by underlying genetic polymorphisms may lead to impaired responses and would, therefore, be associated with adverse outcomes. During the 2009 H1N1 pandemic, several immunogenetic determinants of severe disease were identified. When a novel strain of influenza emerges, the preexisting antibody response directed largely at the surface glycoproteins is rendered ineffective. In these cases, the mechanisms underlying heterosubtypic cross-protection assume a dominant role and it is, therefore, not surprising that immune dysfunction caused by underlying genetic polymorphisms may lead to impaired responses and would therefore be associated with adverse outcomes. During the 2009 H1N1 pandemic, several immunogenetic determinants of severe disease were identified. Allele. The CCR5 protein is a chemokine receptor expressed primarily on T cells, macrophages, and dendritic cells. CCR5 plays a pivotal role in mediating leukocyte chemotaxis in response to chemokines (including RANTES, MIP-1α, and MIP-1β) and is believed to be important in the homing of many immune cell subsets, including regulatory T cells and Th17 cells, to mucosal surfaces. Until recently, the purported role of CCR5 in supporting the antiviral immune response was limited to appreciation of the effect of receptor deficiency in protecting from HIV infection and disease progression among individuals homozygous for the Δ32 allele. The understanding of the roles played by CCR5 was expanded when the Δ32 allele was found to be associated with an increased risk of symptomatic and fatal West Nile Virus (WNV) infection [20] [21] [22] , a severe adverse reaction to the live yellow fever virus vaccine, and with severe tick-borne encephalitis symptoms [23, 24] . Together, these data suggest that CCR5 may also play a critical role in the immune response to flavivirus infections. The spectrum of symptomatic severity observed during the 2009 H1N1 pandemic led our group to study CCR5 genotype among patients requiring intensive care admission and respiratory support for severe H1N1 symptoms. Among twenty samples of confirmed severe pH1N1 infection, the CCR5Δ32 allele was found in 5 out of 9 of the Caucasian individuals, giving a Caucasian allele frequency of 27.8% [25] (Table 1 ). This observed frequency is approximately 2.5 Polymorphisms previously linked to IgG2 deficiency, but not corroborated in H1N1 patients [28] [29] [30] NLRP3 Association with dysregulation of inflammatory response (2107), alteration of NLRP3 mRNA stability and enhancer activity [33, 34] HLA Various alleles Influenza-specific CTL responses exhibit varying frequency and magnitude across various HLA alleles [35] times higher than that reported for local North American Caucasian populations [26, 27] . Given the small sample size available in this cohort, further studies will be required to conclusively determine the impact of CCR5 deficiency on pH1N1 susceptibility and severity. Similarly to IgG1, IgG2a and IgG2b are able to bind to Fc receptors with high affinity and are thought to be important in protecting against influenza infection. A group of Australian investigators identified an index case of severe influenza in a pregnant woman with IgG2 subclass deficiency and subsequently measured total IgG and IgG subclasses in all patients with pH1N1 infection requiring ICU care (many of whom were pregnant) compared to less severe controls and asymptomatic pregnant women presenting to antenatal clinic. A low level of IgG2 was correlated with severe pH1N1 infection after multivariate analysis. Measurement of IgG2 after 90 days among 15 of the surviving IgG2-deficient patients showed that 11 remained IgG2 deficient despite albumin levels returning to baseline values [28] . Additionally, a case-control study from China enrolled 38 Asian patients with respiratory failure due to severe pandemic influenza and compared IgG2 levels with 36 mild cases. They did not find any cases of selective IgG2 deficiency, but did observe significantly lower levels of IgG2 among the severe cases (despite normal levels of the other IgG subclasses) [29] . The authors looked for the presence of FcγRIIa and IGHG2 genotypes ( Table 1 ) that were previously shown to be associated with IgG2 deficiency, but found similar rates among cases and controls. They did, however, corroborate the previously reported finding [30] of cytokine dysregulation among severe cases of infection and suggested that the mechanism responsible for the low IgG2 is the more robust Th1 response and a suppressed Th2 response. Given the lack of FcγRIIa and IGHG2 genotype data available from the Gordon et al. study [28] , the impact of these polymorphisms on IgG2 levels and severe pH1N1 infection remains to be determined. Severe Influenza. Genetic polymorphisms associated with pH1N1 susceptibility and disease severity identified to date are limited, and much of the data is derived from small cohorts. An improved understanding of the sequence of immune responses to influenza as well as the application of newer technologies that employ high-throughput expression array or sequencing technologies can be used to guide a more focused approach to identify specific pathways that may be differentially activated by individuals with severe disease. Based on available data, we can identify several immune pathways and their genetic variants that warrant further investigation. Recently, emphasis has been given to the role of the inflammasome in viral infections and, specifically, influenza. NLRP3 (NOD-like receptor family, pyrin domain-containing 3) inflammasomes are multiprotein complexes containing NLRP3, ASC (or Pycard), and caspase 1. Activation of cytokine/chemokine release through NLRP3 requires a signal derived from Toll-like receptor (TLR) stimulation, with resultant production of pro-IL-1β, -IL-18, and -IL-33. The prointerleukins are in turn cleaved to their respective active forms by caspase 1, which requires the input of an additional signal. The second signal in the context of influenza has been elegantly demonstrated by Ichinohe et al. [31] , who showed that golgi-localized H1N1 M2 is both necessary and sufficient to trigger inflammasome activation. Differential expression of the components of the two signaling cascades that are required for inflammasome activation may therefore explain differences in influenza disease severity. Indeed, mouse knockout studies have shown that intact inflammasomes are necessary for innate immune responses to influenza A, chemokine production, and late stage viral clearance (reviewed in [32] ). Evidence suggests that NLRP3-mediated signaling is also important in cellular recruitment and tissue repair during infection. A similar requirement for proper ASC function was observed in adaptive influenza immune responses. Multiple NLRP3 SNPs have been associated with dysregulated inflammation responses and NLRP3 mRNA stability in humans but have not been examined in the context of pH1N1 susceptibility or mortality [33, 34] (Table 1 ). The CD8+ T-cell response is a strong predictor of vaccine-induced protection and is thought to be particularly valuable in the elderly. Because this response is focused on more conserved viral proteins, it has the additional benefit of providing some cross-reaction with new influenza strains [46, 47] . Undoubtedly, multiple factors underlie the differences in disease severity among ethnic groups, as previously discussed. From an immunogenetic perspective, however, HLA alleles are among the most variable human genes, and it is therefore conceivable that variable proportions of HLA class I alleles among ethnic groups may lead to qualitatively and quantitatively distinct CD8+ T-cell responses, as well as differences in immunodominant epitopes (Table 1) . Boon et al. [35] have demonstrated that the frequency of CTL responses specific for the HLA-B8restricted epitope NP 380-388 was lower in HLA-B27-positive donors than in HLA-B27-negative donors. They also showed that the HLA-A1-restricted epitope NP 44-52 responses were higher in HLA-A1-, -A2-, -B8-, and -B35-positive donors than in other donors. These observations suggest that the epitope specificity and magnitude of the CTL response is related to the HLA class I genetic background [35] . The role of cell-mediated immunity in ameliorating infection caused by novel influenza strains has been the focus of intense study [48] and it is, therefore, the most compelling area to investigate in order to identify immunogenetic factors that predict severity of pandemic H1N1 influenza. A comprehensive investigation was undertaken by Bermejo-Martin et al. [49] in a study from Spain. They enrolled 19 critically ill patients with primary pH1N1 influenza pneumonia and used gene expression analysis in order to identify host immune responses associated with severe disease defined by illness requiring mechanical ventilation. They identified impaired expression of a number of MHC class II and MHC class I genes, T-cell receptor-associated genes, and also of a cluster of genes thought to be involved in dendritic cell maturation, indicating defective antigen presentation in the most severe group of patients. They found further evidence for the effect of altered antigen presentation on the development of an appropriate adaptive response against the virus in the impaired expression of a group of genes critical to the activation and function of both T and B cells. The group with severe illness also showed higher expression of genes involved in IL-6 and IL-10 pathways, and these results were in concordance with the high serum levels of IL-6 and IL-10 in the group dependant on mechanical ventilation. The authors concluded that severe disease is associated with an impaired transition from innate to adaptive immunity in response to the pH1N1 virus, similar to observations in the context of SARS and severe infections caused by H5N1. The impaired adaptive response was also associated with delayed viral clearance. This study did not, however, explore the role of genetic polymorphisms in this immune dysregulation. Pathogenic avian influenza A/H5N1 viruses are endemic among poultry populations across Asia and Africa and present an ongoing risk for avian-to-human transmission. As of June 22, 2011, the WHO reports a total of 562 confirmed H5N1 cases and 329 deaths across 15 countries worldwide [50] . Although human-to-human transmission of H5N1 has so far been rare, the potential for viral evolution into a more transmissible strain raises the possibility that H5N1viruses could cause a pandemic. Given the high mortality rate currently associated with human H5N1 infection, a thorough understanding of the immune response and the underlying mechanisms of viral pathogenesis is crucial to improve treatment and to identify highly susceptible populations. To date, much of the research into the immunobiology and pathogenesis of human H5N1 infection has focused on the H5 haemagglutinin protein and the impact of viral genetic polymorphisms on viral pathogenicity. Although studies have begun to characterize of the role of the host immune response in pathogenesis, the impact of genetic variability on susceptibility and disease severity remains an important gap in our current knowledge. By identifying host genetic polymorphisms that exacerbate immunopathology or provide protection, we will be able to improve treatment and future vaccines [1, 51] . Severity. The precise impact of host genetic variability on H5N1 susceptibility remains somewhat controversial, given the limited case data available and the relatively low number of published studies. A case study in Indonesia [52] found evidence of clusters of H5N1 infection among blood relatives that may be indicative of shared genetic susceptibility, but the authors were unable to rule out a shared viral exposure or altered viral pathogenesis in any of the clusters. Similar observations in a number of additional studies have prompted several authors to suggest a potentially strong genetic basis for H5N1 susceptibility [53] [54] [55] [56] [57] . A compilation of confirmed H5N1 cases worldwide found that, on average, 22% of cases occurred in clusters, and only 6% of cases within the clusters were not genetically related to other cluster members [1] . While this data does not conclusively point to genetic variation as an important determinant of susceptibility, it is important to note that human-to-human transmission of H5N1 is very rare, and therefore does not likely explain the high degree of genetic relatedness among cases [1] . It would also be expected that clusters of nonrelated individuals working in the poultry industry would be more prominent than genetically related clusters, if people are at equal risk of infection [1] . Some studies, however, have suggested that the observed clustering of infections among families could occur due to chance alone at the low rates of infection that are observed in H5N1 and highlight the difficulty in drawing conclusions from the currently available data [58] . TLR3 908T/C Missense mutation identified in a patient with influenza-associated encephalopathy [45] host/virus interactions and the qualities of a protective immune response. To date, a number of candidate genes have been identified from both human and mouse immunobiology studies. Mouse models of influenza infection have the advantage of being able to dissect gene expression kinetics and the characteristics of the immune response at various stages of infection. Comparison of infection across inbred mouse lines demonstrates significant differences in viral titre and core temperature, as well as distinct patterns of immune gene upregulation, suggesting an important contribution of host genetic background [59] . Interactions. Genetic variation affecting host proteins required for viral entry and pathogenesis may partially explain the sporadic and rare nature of avian-tohuman H5N1 transmission. Although humans do express the SAα-2,3Gal molecules that are efficiently bound by avian H5N1, their expression is usually limited to the lower respiratory tract and has only occasionally been detected in the nasal mucosa and upper respiratory tract [60, 61] . Additionally, the binding of several influenza strains to human erythrocytes is highly variable, with up to a 40fold difference between individuals tested in one study, suggesting a role for genetic polymorphism in regulating susceptibility [62] . Whether variation in the human ST3 beta-galactosamide alpha-2,3-sialyltranferase 1 (ST3GAL1) gene that produces the SAα-2,3Gal linkage affects H5N1 susceptibility is not known, but remains a possibility [63] . Interestingly, a population genetics study designed to detect human SNPs under virus-driven selective pressure found a significant enrichment of glycan biosynthesis gene SNPs associated with viral selection, including rs3758105 (intronic A/G SNP) in the ST3GAL1 gene [64] (Table 2) . Data demonstrating the infection of upper respiratory tract cells with H5N1 in vitro also suggests the presence of additional cellular receptors for the virus [61] . Alternately, prevention of viral attachment in the respiratory tract is accomplished by host proteins that can sterically hinder viral HA binding, or aggregate and opsonize the virus. These proteins include serum mannose-binding lectin 2 (MBL2) and surfactant, pulmonary-associated protein A1 and D (SFTPA1 and SFTPD, resp.). A SNP in MBL2 (230G/A), resulting in low serum MBL levels, is associated with SARS susceptibility [36, 37] , while polymorphisms in SFTPA1 and SFTPD are associated with other respiratory illnesses [65, 66] (Table 2 ). To date, none of these polymorphisms have been investigated with respect to H5N1 susceptibility. Induction of an innate immune response following infection can occur as a result of the activation of pattern recognition receptors (PRRs), commonly known as Toll-like receptors (TLRs). TLRs recognize many elements of foreign pathogens, including LPS, flagellin, and dsRNA, and initiate signaling cascades that result in the production of type I interferons. Accumulating data suggests that genetic variation in TLRs and their associated signaling components modulates the response to TLR ligands and, consequently, the inflammatory immune response [67] . TLR3 is constitutively expressed on lung alveolar and bronchial epithelial cells and has been shown to contribute to the secretion of multiple cytokines following influenza A infection [68] . Given the data suggesting that H5N1 pathogenicity is due in part to alterations in innate immune responses and hypercytokinemia, it is plausible that polymorphisms altering TLR function could contribute to susceptibility or protection from infection. This hypothesis is supported by a genetic study of a case of influenza-associated encephalopathy, a condition associated with apoptosis and hypercytokinemia [45] . In this case, a missense mutation (908T/C) in the TLR3 gene was identified and was shown to be a loss-of-function mutation, suggesting a protective role for TLR3 signaling in severe influenza infection [45] ( Table 2) . Although these results are consistent with studies suggesting a protective effect of TLR3 in West Nile infection [69] , they are at odds with TLR3 null mouse studies, which have shown reduced proinflammatory cytokine production following cellular stimulation [70] [71] [72] . Consequently, the contribution of TLR genetic variants to H5N1 inflammatory responses remains to be resolved. Induction of the type I interferon response during influenza infection appears to be important in both human and mouse models, as evidenced by the increased expression of genes including Irf1, Ifi202, Oas1, and Mx1 in mouse microarray studies [73] [74] [75] (reviewed in [76] ). This is consistent with the observation that viral evasion and attenuation of the IFN pathway contributes to H5N1 pathogenesis in humans and suggests potential targets for genetic studies [75, 77] . A strong target for analysis includes the myxovirus resistance (Mx) gene, which encodes interferon-induced antiviral proteins that inhibit viral RNA transcription and consequently confer influenza resistance in mouse lines with functional Mx1 alleles [78] . Polymorphisms in swine Mx genes have also been associated with influenza susceptibility [79] and multiple SNPs in the human MxA gene have been associated with variability in IFN responsiveness in Hepatitis C infection [80] and SARS susceptibility [38] [39] [40] (Table 2) . Specifically, the −123C/A promoter SNP associated with SARS protection correlates with increased basal MxA expression, leading the authors to speculate that −123 genotype may be an important determinant of H5N1 susceptibility [39] . Although human MxA protein has been shown to inhibit influenza replication [81] , no studies have looked for an association between MxA SNPs and H5N1 disease outcome or susceptibility. Because MxA is located on chromosome 21, studies have compared susceptibility to respiratory infections between wild-type and trisomy 21 patients (who exhibit increased MxA expression), but found greater susceptibility among the trisomy 21 group [82] . This group of patients is known to suffer from a multilevel T-cell dysfunction, however, making the role of MxA in respiratory immunity somewhat unclear. Polymorphisms in OAS1 (2 ,5 -oligoadenylate synthetase 1; an interferon-induced antiviral protein) have also been associated with SARS susceptibility and progression [38, 40] and West Nile infection [41] . Consistent with the idea of increased H5N1 susceptibility and pathogenesis associated with poor IFN responses, the OAS1 SNP rs1077467 is correlated with reduced OAS-1 protein activity and associated with increased susceptibility to West Nile infection and in vitro viral replication [41] . Comparison of severe H5N1 infections with uncomplicated seasonal influenza infections revealed a pattern of increased viral load and elevated cytokine production in the respiratory tract and serum [75, 83] . The robust cytokine/chemokine response often seen in H5N1 infected patients (hypercytokinemia) is believed to be at least partially responsible for the observed pathogenesis and high fatality of H5N1 infection. Elevated cytokines both in vivo and in vitro include IFNγ, sIL-2R, IL-6, IP-10, TNFα, and MCP-1 [75, [84] [85] [86] . Mouse models of H5N1 infection also demonstrate elevated levels of MCP-1, MIP-1α, IL-6, and IFNγ, even compared to 1918 H1N1 virus infection [87] . Knocking out IL-1R in mice exacerbates H5N1 pathology and suggests that IL-1β-mediated signalling may be important in protection [75] . In ferret models, IP-10 upregulation and signaling through CXCR3 was determined to be a major component of H5N1 disease severity and mortality [88] . Expression of many of these chemokines and cytokines in humans is modulated by SNPs in their promoter regions, including MCP-1 −2518 G/A [89] , IP-10 −201G/A [90] , and IL-6 −174G/C [91] . Genetic variants affecting expression and function of chemokine receptors may also modulate influenza pathogenesis, as CCR5 knock-out mice exhibit increased influenza mortality, whereas CCR2 knockout strains show increased survival ( Table 2) ; both of these effects appear to be related to the kinetics and strength of macrophage recruitment to the lung [42] . In vitro evidence further suggests upregulation of CCR5 on monocyte-derived macrophages that may enhance pathogenesis [92] . Although relatively few studies have systematically evaluated the influence of genetic polymorphisms on susceptibility and disease severity in zoonotic H1N1 and H5N1 infections, the data available suggest that host immunogenetic variation could play an important role in determining the outcome of the immune response. With improvements in surveillance and case confirmation as well as new sequencing and gene expression platforms, we now have the capability to study host genetic variants among severe respiratory illness cases. Although several challenges to conducting such a study include ethical permission to carry out genetic polymorphism studies, the need for large numbers of wellcharacterised clinical specimens with relevant clinical data, difficulty to obtain sufficient number of samples from severe and fatal cases at a single institution, and difficulty in identifying mild controls. The extreme cases of human H5N1 disease are very rare, sporadic, with scattered cases in different countries, adding economic and political sensitivities associated with this disease. Overcoming these barriers and conducting collaborative research can lead to insights that will shed light on the varying degree of susceptibility observed between populations during the recent H1N1 pandemic and will provide greater insight into the hostpathogen interactions that determine disease course during severe H5N1 infection.

In China, Students in Crowded Dormitories with a Low Ventilation Rate Have More Common Colds: Evidence for Airborne Transmission

OBJECTIVE: To test whether the incidence of common colds among college students in China is associated with ventilation rates and crowdedness in dormitories. METHODS: In Phase I of the study, a cross-sectional study, 3712 students living in 1569 dorm rooms in 13 buildings responded to a questionnaire about incidence and duration of common colds in the previous 12 months. In Phase II, air temperature, relative humidity and CO(2) concentration were measured for 24 hours in 238 dorm rooms in 13 buildings, during both summer and winter. Out-to indoor air flow rates at night were calculated based on measured CO(2) concentrations. RESULTS: In Phase I, 10% of college students reported an incidence of more than 6 common colds in the previous 12 months, and 15% reported that each infection usually lasted for more than 2 weeks. Students in 6-person dorm rooms were about 2 times as likely to have an incidence of common colds ≥6 times per year and a duration ≥2 weeks, compared to students in 3-person rooms. In Phase II, 90% of the measured dorm rooms had an out-to indoor air flow rate less than the Chinese standard of 8.3 L/s per person during the heating season. There was a dose-response relationship between out-to indoor air flow rate per person in dorm rooms and the proportion of occupants with annual common cold infections ≥6 times. A mean ventilation rate of 5 L/(s•person) in dorm buildings was associated with 5% of self reported common cold ≥6 times, compared to 35% at 1 L/(s•person). CONCLUSION: Crowded dormitories with low out-to indoor airflow rates are associated with more respiratory infections among college students.

''Common cold'' is a conventional term for a mild upper respiratory illness, with symptoms such as nasal blockage and discharge, sneezing, sore throat and cough [1] . Adults typically have 2-5 common colds per year, and children 4-8 colds [2] . Although such infections are often regarded as trivial, the cost to society is large [3] . Rhinoviruses have been associated with 40-65% of ''common colds'' through the year [4] , and up to 80-92% of colds during outbreaks [5] . Cross-infection from an infected person to a healthy person depends on a number of factors, including how many viral particles are shed by the infected person, and the viral particles' survivability, both over time and with respect to distance from source in a shared environment. Three main mechanisms have been proposed for transmission of viruses causing airways infections: N contact with secretions that contain the virus, either directly (e.g. hand to hand) from an infected person or indirectly from surfaces (e.g. door knob), N ''large'' airborne droplets, which are produced by an infected person during talking, sneezing, or coughing, and can only spread in air for a distance of less than 1-2 m before falling down, N ''small'' droplet nuclei (dried droplets), that can stay airborne for an extended time and be transported long distances. Despite many years of study, the routes of spread of viral airways infections remain controversial. One opinion is that the virus is transferred through direct contact [6] , while the other is that the virus is transferred through airborne spread [7, 8] . During the SARS epidemic, early preventive messages to the public were to wash hands and, generally to avoid ''direct'' contact spread. Later, analysis of the temporal and spatial distributions of SARS cases in a large community outbreak in Hong Kong and the correlation of these data with the three-dimensional spread of a virus-laden aerosol plume indicated an important role for airborne spread of droplet nuclei [9] . The influence of building characteristics including ventilation on the spread of viral respiratory infections has begun to receive increased attention from the public, government, media and scientists [10] . Brundage et al. [11] studied the risk of febrile acute respiratory diseases at four army training centers and found that disease rates were significantly higher among trainees in modern energy efficient barracks that had a low ventilation rate. Menzies et al. [12] suggested that there was a relationship between lower ventilation rates and more frequent tuberculosis infections among hospital workers. Milton et al. [13] reported an association between sick leave of employees and outdoor air supply rate. Myatt's [14] study showed that the probability of detecting airborne rhinoviruses was positively associated with weekly average CO 2 concentration in an office. Other factors found to be associated with rate of infectious diseases include occupancy level [15] , cleaning routines and ''damp'' buildings [16] . With respect to crowding, direct and surface contact as well as airborne transmission both appears to be factors in disease transmission. Hoge et al. found that severe overcrowding and inadequate ventilation contributed to an outbreak of pneumococcal disease in a large urban jail [17] . In China, one 20 m 2 dormitory room is shared by 6-8 bachelor students or 4 master students or 3 PhD students. While such crowded spaces may be important sites for the propagation of respiratory infections, few studies have examined dorm room ventilation and its possible association with infection transmission. The aim of this paper is to test whether the common cold is associated with how crowded a dorm room is and how well ventilated it is among college students in China. Verbal consents were obtained from participants, since participants did not want to be tracked back by signature. Both the study and the consent procedure were approved by the ethics committee at Tianjin University. This study is part of the ''Dorm Environment and Occupants' Health'' study, which was carried out from 2006 to 2007 at Tianjin University, China. Details of the recruitment process and questionnaire contents have been previously described [18] . In brief, this study consisted of two phases. In Phase I, demographic information, the health status of 6500 students, and building and room characteristics of 2117 dorm rooms at Tianjin University were surveyed by questionnaires. The questionnaire survey was anonymous, but building number and room number were reported by participants. Project members visited dorm rooms, distributed questionnaires and explained to participants how to fill out questionnaires. The questionnaires were collected 2 days later. The questions on common cold infections were ''how many times have you had a common cold in the previous 12 months (options: ,6 times; 6-10 times; .10 times)'' and ''how long does a common cold usually last (options: ,2 weeks; 2-4 weeks; .4 weeks)''. Other questions were about frequencies of window opening, cleaning routines and environmental tobacco smoke (ETS) exposure. In Phase II, air temperature, relative humidity and CO 2 concentration in dorm rooms were measured by indoor air quality monitor PS 31 (http://www.sensotron.pl) for 24 hours. Air quality monitors were calibrated at the International Center for Indoor Environment and Energy, Technical University of Denmark prior to measurements. Dorm occupants reported opening status of doors and windows at day and at night during measurement The out-to indoor air flow rate at night was calculated from an analysis of the build-up period of metabolic CO 2 produced by sleeping occupants (1:00 a.m.-8:00 a.m.) [19] . Calculation details are described in Information S1. CO 2 concentrations of dorm rooms were measured both in the summer (May-Jul., 2006) and in the winter (Dec., 2006-Apr., 2007) [20] . The average indoor air temperature and relative humidity at night were calculated (1:00 a.m.-8:00 a.m.). Outdoor CO 2 concentration and meteorological parameters were also measured on campus during the same time. The associations among gender, age, whether family member ever had asthma and allergy, environmental tobacco smoke, cleaning routine, window opening frequency, occupancy levels, and self-reported common cold incidence and duration were analyzed by Chi-square tests. Adjusted odds ratios of crowdedness and air flow rate for common cold infections were evaluated in multiple logistic regression models. A carbon dioxide-based risk equation [21] was used to calculate the basic reproductive number of common colds which was compared to the self-reported infection rate. A P value less than 0.05 indicates statistical significance. SPSS software 15.0 was used to perform the statistical analyses. In Phase I, 3712 students living in 1569 dorm rooms in 13 buildings answered the questionnaire, giving a response rate of 57%. Surveys for 276 students were excluded from the analysis due to missing information. Forty eight percent (48%) of students were female. PhD students' mean age was 29 years, master students 25 years and bachelor students 22 years. Monday through Friday, 18% of participants spent less than 2 hours indoors watching TV/playing games per day, 36% spent 2-10 hours per day, and 46% spent more than 10 hours per day. Dorm buildings had 3-12 floors, with 26-43 dorm rooms per floor. All floors in each dorm building are homogeneous with regard to occupants' gender and education level. Dorm rooms consisted of one simple bedroom. Each floor provided two washing rooms and restrooms. Six bachelor students, 4 master students or 3 PhD students shared one dorm room with a volume of 50-70 m 3 . The average density was 5 m 2 per person. Based on the questionnaire data from Phase I, 238 dorm rooms with 473 students living in these dorms were evaluated for Phase II. The evaluated dorm rooms represented different building structures, construction periods, locations and occupancy levels. There were no significant differences in students' ages, gender, self-reported common cold incidence or duration between Phase I and Phase II. In the questionnaire survey of Phase I, 249 out of 3436 (7.3%) students reported 6-10 common colds in the previous 12 months, while 94 (2.8%) reported more than 10 common colds. Four hundred and thirty six (12.8%) students had common colds lasting for 2-4 weeks, while 65 (1.9%) reported colds lasting more than 4 weeks. Demographic information and living habits of dormitory occupants, and their associations with common cold are summarized in Table 1 . Atopy was associated with increased incidence and longer duration of common cold. Male students were more susceptible than females, but had shorter duration colds. Females cleaned rooms more often than males, cleaning rooms at least twice per week 52% compared to 31% for males, and smoked less (1.4% vs. 15.9%). Passive smoking had a significant effect on the incidence of common cold (p = 0.029), but after adjustment for environmental tobacco smoke, males were still at greater risk for common colds (p = 0.010). Younger students lived in more crowded rooms and reported longer duration colds. Crowding, not age, was shown by stratification for occupancy level to be the significant association with common cold duration. Self-reported common cold incidence and duration are compared for different occupancy levels in Figure 1 . With incrementally increasing occupancy in dorm rooms, the proportion of occupants with $6 common colds increased significantly (p = 0.002), as did the proportion of occupants with $2 weeks common cold duration (p = 0.000). The odds ratios of crowdedness for common cold incidence of $6 times and duration of $2 weeks, adjusted for gender, age, hours spent indoors, family members' asthma and allergy history, environmental tobacco smoke exposure are shown in Figure 2 . Students in 6-person rooms were about 2.0 times as likely to have a common cold incidence $6 times per year and a duration $2 weeks, as students in 3-person dorm rooms. For Phase II, the evaluated dorm rooms were located in 13 buildings. Four were built between 1940 and 1960, two between 1977 and 1983, three between 1993 and 1999, and four after 2000. For newly constructed dorm buildings, concrete structure and PVC frame windows were used instead of the brick-stone structure and the wooden frame windows used in older buildings. Ventilation for all dorm rooms consisted solely of opening doors and windows. The out-to indoor air flow rates for rooms measured during summer varied significantly, from 0.8 to 110 L/s per person, with a median of 18 L/s per person. Air flow rates measured in the heating season (from Dec. 5, 2006 to Apr. 14, 2007) varied from 0.3 to 24 L/s per person, with a median of 3.0 L/s per person. Ninety percent of the dorm rooms had an outto indoor air flow rate less than 8.3 L/s per person. The average indoor air temperature (mean 28.0uC, 95% confidence interval (CI) 27.8uC-28.3uC, range 22.0uC-32.1uC) and relative humidity (mean 54%, 95% CI 53%-55%, range 27%-78%) in summer were high and had large variations consequent to opening doors and windows as the sole mode of ventilation. During the winter season when the heating system was in use and doors and windows were closed, weather conditions had less influence on the indoor thermal environment (temperature: mean 21.0uC, 95% CI 20.7uC-21.3uC, range 15.4uC-26.5uC; relative humidity: mean 40%, 95% CI 38%-41%, range 18%-72%). Data for temperature and relative humidity in rooms with different occupancy levels and out-to indoor air flow rates are shown in Table 2 . In summer, relative humidity and temperature were not different in rooms with different air flow rates. An inverse association between occupancy level and relative humidity was caused by the measurement sequence (6-person dorms were measured at the driest time in May, whereas 3-person dorm rooms were measured in July when outdoor relative humidity was higher). Outdoor climate is the dominating factor in determining the indoor temperature and relative humidity in summer. In winter, rooms shared by 6 people had the highest relative humidity and temperature at night. A low out-to indoor air flow rate was related to a significantly higher relative humidity (p = 0.000). However, common cold infections were not significantly associated with indoor air temperature (p = 0.806) and relative humidity (p = 0.642). Figure 3 shows that the lowest quartile of out-to indoor air flow rates per person in both summer and winter were associated with an increased proportion of occupants with $6 common colds in the previous 12 months. The adjusted odds ratios of ventilation rates for common cold infections increased slightly across the quartiles. The critical ventilation rate, below which common cold incidence increases, is identified. When ventilation rate is below 6 L/s per person, the common cold incidence in dorm rooms with average 4 occupants increased from 10% to 12%. When ventilation rate is below 1 L/s per person, the common cold incidence increased from 10% to 15%. In our study, old buildings had more dampness problems, while new buildings using modern construction technologies had smaller ventilation rates [21] . Dampness problems have been reported to be associated with an increased incidence of common cold infections [18] . In order to eliminate the influence of indoor environmental factors other than poor ventilation, the mean ventilation rates in newly constructed dorm buildings were calculated and related to the percentage of occupants with common cold infections more than 6 times annually. The ventilation rates in winter are less than those in summer, and may help nail down the critical ventilation rate, below which common cold incidence increases. Figure 4 shows that the infection rate of common colds in the ''tight'' buildings constructed after 1993 is, in winter, associated with mean ventilation rate. There were 7 buildings constructed after 1993. One building was not included in the analysis because measurements were performed in only 9 dorm rooms. On average, there were 1140 occupants in each dorm building. A mean ventilation rate of 5 L/ (sNperson) was associated with $6 common colds per year in 5% of occupants , compared to a 35% for 1 L/(sNperson). There were 6 buildings constructed before 1993, among which 4 buildings had ,10 dorm rooms measured in winter and were excluded from the analysis. Of the remaining 2 buildings, one had mean ventilation rate of 5.7 L/(sNperson) and a common cold infection rate of 23.8%, while the other had a mean ventilation rate of 6.4 L/ (sNperson) and a common cold infection rate of 7.1%. The Wells-Riley equation estimates the number of secondary infections that arise when a single infectious case is introduced into a population where everyone is susceptible [22] . This number is called the basic reproduction number. Rudnich and Milton [23] expanded the Wells-Riley equation to apply to situations with nonsteady state conditions and variable ventilation rates: Where R A0 is the basic reproduction number; n is the number of occupants; f is the re-breathed fraction; and I is the number of infectors ( = 1). q is the quantum generation rate by an infected person (quanta/h), where a quantum is the amount of infectious material needed to produce infection in 63% of uniformly exposed animals, and is therefore 1.25 times the median infectious dose, 1.256ID 50 . t is the exposure time (h); f = (C-C 0 )/C a , where C a is the volume fraction of CO 2 added to exhaled breath, C is the volume fraction of CO 2 in indoor air, and C 0 is the volume fraction of CO 2 in outdoor air. The incidences (,6 times; 6-10 times; .10 times) and durations (,2 weeks; 2-4 weeks; .4 weeks) of common colds in the previous 12 months for different occupancy levels (6-people; 4people; 3-people per dorm) were self-reported by occupants. The mean duration of a common cold is 7-10 days [1] . For this study we assumed that the duration of a common cold was 9 days. Although many viruses can produce symptoms of common cold, rhinovirus is the most frequent cause of the common cold [24] . Riley and Nardell suggested that q for rhinovirus is in the range of 1-10/h [25] . Here we inferred q = 9/h. We assumed that the infector remained in the dorm room 8 hours per day. The average CO 2 concentrations in each dorm room from 1:00 a.m. to 8:00 a.m. were calculated. The estimated and self-reported number of common colds in each day in winter is compared (Table 3) . These two numbers fit very well indicating the validity of this CO 2 -based risk model in predicting infection rate of infectious disease like common cold. If for a given population and infectious agent, the basic reproductive number .1 then that agent can spread in the population. The critical re-breathed fraction (f c ), corresponding to a basic reproduction number of 1, can be derived from Equation (1), In the present study, the critical re-breathed fraction in rooms with different occupancy levels and the associated critical indoor CO 2 concentrations above background (outdoor CO 2 concentration) were calculated from Equation (2), both as a function of exposure time ( Figure 5(a) ) and quantum generation rate ( Figure 5(b) ). Thus Figure 5 predicts the critical indoor CO 2 concentrations beyond which infectious disease will spread. The family of curves in Figure 5 (a) describes the trends of the critical indoor CO 2 concentrations above outdoor values (C-C 0 ) as a function of exposure times for risk of respiratory infections. The quantum generation rate used was 2/h. The critical CO 2 concentration above the background levels off if the common cold lasts more than 3 weeks (exposure time 8 hours/day, totally 168 hours) (Fig 5(a) ). This indicates that even for less infectious agents with quanta generation rate no more than 2/h, a full fresh outdoor air system without recirculation of indoor air needs to be used in environments where people spend extended time (for example bedrooms, dorms, schools, daycare centers) in order to prevent viral infections. In Figure 5 (b), the exposure time was assumed to be 56 hours (8 hours/day, i.e. 7 days). It shows that the current ASHRAE standard of 700 ppm above the background level [26] would not prevent the infection from being spread in a dorm room with 6 occupants unless the quantum generation rate of infectious agents is no more than 1 quantum/h (Fig. 5(b) ). The campus living style and dormitory conditions of students at Tianjin University is typical of China. The sample size in our study is large, and the response rate was reasonably good (57%). No significant difference was found between respondents and nonrespondents in reporting wheeze and dorm room dampness [18] . Thus it is highly unlikely that selection bias impacted the findings of this study. Common cold is a conventional term for a mild upper respiratory illness. College students can be expected to understand what ''common cold'' refers to. There is no evidence to suggest that bachelor students have a different memory in reporting common cold infection, compared to PhD students. Compared to home environment, dorm buildings are perceived to be very crowded no matter whether 3 or 4 or 6 people share a 20 m 2 room. Even students in 3-people-shared dormitory think their space is crowded. Therefore, the significant association between occupancy level and incidence of common colds, and the dose-response relationship between ventilation rate and incidence of common colds cannot be explained by reporting bias. The occupants' education level was not adjusted for when calculating the odds ratios of crowdedness for common cold infections since 3 PhD students or 4 master students or 6 bachelor students share one dorm room with similar volume. Education level itself should not be a confounding factor. Psychological stress, related to education status may have effect on common cold as indicated in a previous study [27] . However, our study found that less crowded dorm rooms occupied by PhD students were associated with less common cold infections. This cannot be explained by psychological stress since PhD students are supposed to be more stressed than master or bachelor students. The summer measurement was from May to July and winter measurement from December to April. In summer measurements, 6-people-shared dormitories were measured first, followed by 4 or 3 people shared dormitories. In winter measurement, dorm Figure 3 . Associations between ventilation rate and common cold annual incidence $6 times. 1 Proportion of occupants with $6 common colds in the previous 12 months. 2 Odds ratios were adjusted for gender, age, family member allergy history, exposure to environmental tobacco smoke, building age and crowdedness. AOR: adjusted odds ratio; CI: confidence interval. doi:10.1371/journal.pone.0027140.g003 buildings were measured randomly. There could be a potential systematic bias for summer measurement, but not for winter measurements. During the measurements, outdoor CO 2 concentrations and meteorological parameters were monitored. In principle, air change rate in buildings with natural ventilation system is not influenced by air relative humidity. Outdoor air temperature itself and the consequent occupants' behavior (e.g. opening doors/windows) may influence the air change rate in dorm rooms. In our study, the opening of doors/windows was reported by occupants themselves. In winter, occupants tended to close doors and windows tightly, so that variations in winter outdoor temperature had little influence on ventilation rate in D i is the assumed number of common cold infections in winter under different self-reported incidence rate, times. i indicates common cold incidence. i = 1, 2, 3. 1common cold less than 6 times in the previous 12 months; 2-common cold 6-10 times; 3-comon cold more than 10 times. We assume D 1 = 3; D 2 = 6; D 3 = 8. O j is the occupancy level, person/room. j indicates occupancy level. j = 3, 4, 6. 3-three people per dorm room; 4-four people per dorm room; 6-six people per dorm room. D i,j is the proportion of students with different self-reported common cold incidences, %. M is the duration of a common cold, days. We assume M = 9 days [1] . T is days in winter season, 120 days. C j is the average CO 2 concentration from 1:00 a.m. to 8:00 a.m. in rooms with different occupancy levels, ppm. f j is the re-breathed fraction of indoor air in rooms with different occupancy levels. f j = (C j -C 0 )/C a . C a is the volume fraction of CO 2 added to exhaled breath, 37000 ppm. C 0 is the volume fraction of CO 2 in outdoor air, 300 ppm. q is the quantum generation rate by an infected person, quanta/h. We assume q = 9 quanta/h [25] . t is the time a infector remaining in the dorm room, hour/day. We assume t = 8 hours per day. doi:10.1371/journal.pone.0027140.t003 dorm buildings. In summer, the mean outdoor air temperature was 29.6uC, ranging from 22.5uC to 35.2uC. The median air change rate was 4.42 h 21 and 4.67 h 21 when outdoor air temperature is below and above 29.6uC. There was no significant difference of air change rate for different temperatures in summer (p = 0.319). Therefore, it is reasonable to assume that, the air change rates measured in summer and winter are representative for respective season, without influence from small climate changes within each period. While it is possible that some of the self-reported common colds were influenza, the infection rate of flu among adults is approximately once per year in this part of China. Therefore, this possible error would not change our results. Moreover, common colds and influenza are spread in a similar way; the present study could have been titled ''airways infections''. In each dorm room, CO 2 concentrations were measured for 24 hours in both summer and winter. As measurements were made over a long period, i.e. summer measurements between May and July and winter measurements between December and April, and for 238 rooms, the mean values of ventilation rates should be valid for rooms with different occupancies and opening status of windows/ doors, and for changes in the outdoor climate. There were imperfections in our data collection. In some rooms occupants may have had the window open during the night measurements in winter. Perhaps the incidence of common cold was influenced by an influenza epidemic. These sources of error would shift our findings towards the null hypothesis, that there was no association between common cold infections and dorm crowdedness or ventilation rate. Our findings are robust in spite of these possible problems. Thus, it is likely that more measurements and more accurate data on types of airways infections would show an even stronger association. The out-to indoor air flow rate required by the Indoor Air Quality Standard of China is 8.3 L/s per person [28] . In the present study, 90% of the dorm rooms measured during winter had night-time ventilation rates less than this value. CO 2 concentration in corridors was not measured, so that the fresh out-to indoor air flow rate may have been even lower than the calculated value in cases when corridor windows were closed. The suggested dose-response relationship between dorm ventilation rate and common cold infections among occupants can be extrapolated to other crowded public premises with substandard ventilation rate, meaning a possible important public health topic for e.g. schools, daycare centers. Although it is widely held that people in crowded spaces have more airways infections [15, 29] , there are few studies on this. Our study is among the first published suggesting a relationship between occupancy levels, ventilation rates, and respiratory infections. With 6 occupants instead of 3 in a 20 m 2 dorm room, the proportion of occupants with incidence of more than 6 common colds in the previous 12 months doubled. When crowdedness is adjusted for, a lower ventilation rate is associated with an increased risk of common cold. This finding is consistent with Shendell's study in schools, which showed that a 1000 ppm increase in dCO 2 (difference between indoor and outdoor CO 2 levels) was associated with a 0.5%-0.9% decrease in annual average daily attendance [30] . For office buildings, Milton found that short-term sick leave was reduced by 35% at 24 L/s per person compared to 12 L/s per person outdoor air flow [13] . A crucial question is whether the increased frequency of common colds in crowded places is due to direct contact (or via surfaces), via droplets or via droplet nuclei. The strong association with ventilation in this study indicates that airborne transmission is important and perhaps the main route. Crowdedness and outdoor air ventilation per person are important for the spread of airborne infectious diseases in rooms such as dorms where people spend a lot of time. Respiratory viruses can be transmitted through air so that transmission is modulated by outdoor air supply rates. Further studies are warranted. Information S1 Dormitory outdoor air flow rate calculation by using CO 2 method. (DOCX)

Epidemiology and clinical characteristics of hospitalized patients with pandemic influenza A (H1N1) 2009 infections: the effects of bacterial coinfection

BACKGROUND: Numerous reports have described the epidemiological and clinical characteristics of influenza A (H1N1) 2009 infected patients. However, data on the effects of bacterial coinfection on these patients are very scarce. Therefore, this study explores the impact of bacterial coinfection on the clinical and laboratory parameters amongst H1N1 hospitalized patients. FINDINGS: This retrospective study involved hospitalized patients with laboratory-confirmed H1N1 infections (September 2009 to May 2010). Relevant clinical data and the detection of bacterial coinfection from respiratory or sterile site samples were obtained. Multiplex PCR was used to determine the co-existence of other respiratory viruses. Comparison was made between patients with and without bacterial coinfection. The occurrence of coinfection was 34%; 14 (28%) bacterial and only 3 (6%) viral. Mycoplasma pneumoniae (n = 5) was the commonest bacteria followed by Staphylococcus aureus (n = 3). In univariate analysis, clinical factors associated with bacterial coinfection were age > 50 years (p = 0.02), presence of comorbidity (p = 0.04), liver impairment (p = 0.02), development of complications (p = 0.004) and supplemental oxygen requirement (p = 0.02). Leukocytosis (p = 0.02) and neutrophilia (p = 0.004) were higher in bacterial coinfected patients. Multivariate logistic regression analysis revealed that age > 50 years and combined complications were predictive of bacterial coinfection. CONCLUSIONS: Bacterial coinfection is not uncommon in H1N1 infected patients and is more frequently noted in the older aged patients and is associated with higher rates of complications. Also, as adjunct to clinical findings, clinicians need to have a higher index of suspicion if neutrophilia was identified at admission as it may denote bacterial coinfection.

In April 2009, a novel influenza A (H1N1) virus emerged in Mexico and spread rapidly worldwide [1] . By June 11, 2009 nearly 30, 000 cases had been confirmed across 74 countries including Malaysia, prompting World Health Organization to raise its pandemic alert to phase 6 [2] . After the first reported H1N1 case in Malaysia in May 15, 2009, the numbers increased exponentially and as of May 31, 2010 they totaled 14, 821 with 87 deaths [3] . Thereafter, there have been numerous reports describing the epidemiological and clinical characteristics of H1N1 infections. However, studies focusing on the effects of respiratory pathogen coinfection on clinical and laboratory parameters in the H1N1 infected patients are scarce. Clinicians may assume that a single virus type is involved, as laboratory detection involves PCR specifically targeting H1N1. However, bacterial coinfection had been shown to contribute to morbidity and mortality in previous influenza pandemics [4] . Therefore, this study aims to explore the clinical and laboratory characteristics amongst patients hospitalized with laboratory-confirmed pandemic influenza A (H1N1) infection and the effects of bacterial coinfection on these parameters. This retrospective study was conducted from September 2009 to May 2010 at Hospital Sultanah Aminah Johor Bahru (HSAJB). HSAJB is a 989-bedded tertiary referral centre and the government designated hospital for H1N1 testing in Johor State, Malaysia. As the main General Hospital of Johor, its' patient population is reflective of the larger community in Malaysia. During our study period, which coincided with the peak of H1N1 pandemic activity, all patients regardless of whether they were hospitalized or not, who presented with an influenza-like illness (ILI) were tested for H1N1. Consecutive hospitalized patients with laboratory-confirmed H1N1 infections were identified from microbiology laboratory records. Laboratory diagnosis of H1N1 was made using the Centers for Disease Control and Prevention (CDC) real-time reverse transcriptase polymerase chain reaction (RT-PCR) protocol [5] . Relevant clinical data was retrieved from patients' medical records. The presence of bacterial coinfection from respiratory specimens (sputum, tracheal/nasopharyngeal aspirate, bronchoalveolar lavage) or sterile site samples (blood or pleural fluid) taken within 48 hours of admission was recorded. Mycoplasma pneumoniae infection was diagnosed by serology using particle agglutination test (Serodia-Myco II, Fujirebio Inc., Japan). A single titer of ≥ 160 was considered as diagnostic cut-off titer, based on population background study conducted in Malaysia [6, 7] . All samples confirmed H1N1 positive were stored at -80°C for further analysis using multiplex PCR (Seeplex RV Detection, USA) which detects adenovirus, influenza virus A and B, respiratory syncytial virus, parainfluenza types 1, 2 and 3 and human metapneumovirus. Hematological, liver and renal function parameters on admission were recorded. Data was analyzed using SPSS version 17.0.1; comparing patients with and without bacterial coinfection with a P-value < 0.05 (two-tailed) taken as the level of significance. Variables associated with bacterial coinfection in the univariate analysis were then entered into multivariate logistic regression analysis. After excluding 7 patients (5 incomplete data and 2 for presence of nosocomial pneumonia), data of 50 patients was available for analysis ( Table 1 ). The patients age ranged from 7 months to 82 years (median 20.3 years), with 90% patients (45/50) < 50 years. Excluding 6 * Not assessed in children < 3 years (n = 44) ** A patient may have more than one comorbidity or complications ¶ Includes asthma (n-= 9), Chronic obstructive airway disease (n = 1) and bronchiectasis (n = 1) Ψ Includes chronic myeloid leukemia (n = 1), acute myeloid leukemia (n = 1), meningioma of brain (n = 1) ¥ Includes autoimmune haemolytic anaemia (n = 1), idiopathic thrombocytopenia purpura (n = 1) £Includes cardiovascular disease (n = 2), immunosuppressives (n = 2), hypothyroidism (n = 1), stroke (n = 1) ® Includes liver impairment (n = 12), renal impairment (n = 4) septic shock (n = 2) and ARDS (n = 2). Ω Includes bacterial (n = 14), viral (n = 3). The sites for isolation of 9 non-Mycoplasma bacteria: (blood = 2, sputum = 3, nasopharyngeal aspirate = 3, bronchoalveolar lavage = 2) © Established values in our laboratory, Adults: leukocytes 4-11 × 10 9 /L; neutrophils 2-7.5 × 10 9 /L; lymphocyte 1.5-4 × 10 9 /L; Paediatrics: age-dependent ALF: abnormal liver function (n = 41) (raised alanine aminotransferase/ aspartate aminotransferase or both) ARF: abnormal renal function (n = 42) (raised creatinine) pregnancies, 24 patients (48%) had at least one preexisting comorbidity; lung disease being the commonest. The mean duration of symptoms before hospitalization was 4.4 ± 3.08 days (range 1-14 days). Cough (100%) and fever (98%) were the most common symptoms on admission. Twelve patients (24%) had oxygen saturation < 95% at presentation. Pneumonia was diagnosed in 25 patients (50%) based on clinical and radiological findings. All patients received oseltamivir after admission. Twenty-two patients (44%) required oxygen supplementation. Nine cases (18%) were treated at the intensive care unit (ICU); 6 requiring mechanical ventilation. Thirteen patients (26%) developed complications (single or combination); liver impairment (n = 12), renal impairment (n = 4) septic shock (n = 2) and acute respiratory distress syndrome (ARDS) (n = 2). Two (4%) patients died, resulting from septicaemic shock and severe pneumonia respectively. Forty-five patients (90%) had lower respiratory tract specimens sent for bacterial cultures. The 5 patients without these specimens were children who had difficulty in producing respiratory secretions, however, they appeared generally well with no evidence of pneumonia. Blood cultures were performed in 23 patients (46%) and Mycoplasma pneumoniae serology in 27 patients (54%). Of the 50 H1N1 patients, 17 (34%) were coinfected with a second respiratory pathogen; 14 (28%) bacterial and only 3 (6%) viral. Mycoplasma pneumoniae (n = 5) was the commonest bacterial coinfection followed by Staphylococcus aureus (n = 3), Klebsiella pneumoniae (n = 2), Streptococcus pneumoniae (n = 2), Moraxella catarrhalis (n = 1), Pseudomonas aeruginosa (n = 1), Streptococcus pyogenes (n = 1) and Streptococcus agalactiae (n = 1). Two patients had dual infection; M.pneumoniae/S.agalactiae and S.pneumoniae/M.catarrhalis respectively. The sites for isolation of 9 non-Mycoplasma bacteria were blood (2), sputum (3), nasopharyngeal aspirate (3) and bronchoalveolar lavage (2). The 3 virus detected were parainfluenza; these 3 patients presented with influenza-like illness with no deterioration of clinical findings. A comparison between H1N1 patients with and without bacterial coinfection is shown in Table 2 . Although 90% of patients were < 50 years old, bacterial coinfection was more frequent in patients > 50 years (p = 0.02). The presence of underlying comorbidity provided a suitable niche for bacterial coinfection (p = 0.04). Although ICU admissions, mechanical ventilation, renal impairment, mortality and pneumonia were notably higher in patients with bacterial coinfection, they were not statistically significant. Other factors associated with bacterial coinfection in the univariate analysis were development of complications (p = 0.004), liver impairment (p = 0.02) and supplemental oxygen requirement (p = 0.02). Out of the 50 patients, 12 (24%) had leukocytosis and 13 (26%) neutrophilia. Bacterial coinfected patients demonstrated higher rates of leukocytosis (p = 0.02) and neutrophilia (p = 0.004). On the other hand, lymphopenia (n = 31) was notably higher in single viral H1N1 infection. Multivariate analysis revealed that age > 50 (OR 12.577; 95% CI 1-165.24; p = 0.05)) and development of complications (OR 9.01; 95% CI 1.70-47.67; p = 0.01) were predictive of bacterial coinfection. Forty-one patients (82%) received antibiotics, either as empiric or definitive therapy upon admission and 16% prior to admission All patients with bacterial coinfection were treated with antibiotics; significantly higher rates compared to patients without bacterial coinfection (p = 0.05). The bacterial coinfection rate of 28% amongst our H1N1 hospitalized patients was higher compared to other studies [8, 9] . A large laboratory-based study in the United States demonstrated comparable bacterial coinfection rates to our study with similarly very low frequency of viral copathogen detection [10] . Whilst our finding concurred with several studies [1, 8, 9, 11, 12] that showed H1N1 infections having a predilection for younger patients, patients > 50 years had higher risk of bacterial coinfection in our study. Although concurrent bacterial infection was shown to have a major influence on mortality in previous influenza pandemics [4] , its' role in the current H1N1 pandemic is still evolving. Recent postmortem studies amongst fatal H1N1 cases established a link between bacterial lung infections and increased deaths [13] . Whilst an earlier study showed bacterial coinfection not to be a major contributor to severe disease [12] , a more recent study demonstrated otherwise [8] . In our study, patients with bacterial coinfection were found to have higher risk of developing complications. The presence of underlying comorbidity, liver impairment and supplemental oxygen requirement were significantly higher in bacterial coinfected patients in univariate analysis, although these factors were not predictive in multivariate analysis. Unlike S.pneumoniae, S.aureus and S.pyogenes which are repeatedly reported as coinfecting agents [4, 8, 10, 13] , the high rates of M.pneumoniae coinfection was unique to our study. Although hematological parameters have been mentioned in few other studies [8, 9, 12] , to our best knowledge this is the first study that specifically explored the impact of bacterial coinfection on these parameters. CDC recognizes the importance of early empirical antibiotics in H1N1 infected patients who might have concurrent bacterial pneumonia [13] . Our study showed that leukocytosis and neutrophilia were notably higher in bacterial coinfected patients. This finding could alert physicians about the possibility of bacterial coinfection, as clinical diagnosis may be insufficient and bacterial cultures take time. Eighty-two percent of our patients received empiric or definitive antibiotics at some point during admission which was comparable to high rates in a China study [9] . The limitation of our study includes its' retrospective design and a small sample size which was unavoidable, as we were limited by the actual number of cases during the study period and because it was a single centre study. As such, our study was inadequately powered to examine the influence of certain characteristics. Nasopharyngeal aspirates may have questionable pathogenic role, however the 3 patients with positive NPA were treated with appropriate antibiotics as they were felt to be clinically relevant. Mycoplasma serology was not performed in all patients and the request was based upon physicians' discretion, this may have underestimated the actual number of cases. The preadmission antibiotic therapy could underestimate the bacterial coinfection rates. Despite these limitations, we identified bacteria coinfection in 28% of our patients. In conclusion, our study suggests that bacterial coinfection is not uncommon in H1N1 infected patients and laboratory investigations should go beyond establishing a viral cause alone. Bacterial coinfection was more frequently seen in the older age group and was associated with higher rates of complications. As adjunct to clinical findings, clinicians need to have a high index of suspicion if neutrophilia was identified on admission as it may denote bacterial coinfection. A larger scale study will be useful to further confirm our findings.

Clinical Review: Gene-based therapies for ALI/ARDS: where are we now?

Acute lung injury (ALI) and acute respiratory distress syndrome (ARDS) confer substantial morbidity and mortality, and have no specific therapy. The accessibility of the distal lung epithelium via the airway route, and the relatively transient nature of ALI/ARDS, suggest that the disease may be amenable to gene-based therapies. Ongoing advances in our understanding of the pathophysiology of ALI/ARDS have revealed multiple therapeutic targets for gene-based approaches. Strategies to enhance or restore lung epithelial and/or endothelial cell function, to strengthen lung defense mechanisms against injury, to speed clearance of infection and to enhance the repair process following ALI/ARDS have all demonstrated promise in preclinical models. Despite three decades of gene therapy research, however, the clinical potential for gene-based approaches to lung diseases including ALI/ARDS remains to be realized. Multiple barriers to effective pulmonary gene therapy exist, including the pulmonary architecture, pulmonary defense mechanisms against inhaled particles, the immunogenicity of viral vectors and the poor transfection efficiency of nonviral delivery methods. Deficits remain in our knowledge regarding the optimal molecular targets for gene-based approaches. Encouragingly, recent progress in overcoming these barriers offers hope for the successful translation of gene-based approaches for ALI/ARDS to the clinical setting.

Gene-based therapy involves the insertion of genes or smaller nucleic acid sequences into cells and tissues to replace the function of a defective gene, or to alter the production of a specifi c gene product, in order to treat a disease. Gene therapy can be classifi ed into germline and somatic gene therapies. Germline approaches modify the sperm or egg prior to fertilization and confer a stable heritable genetic modifi cation. Somatic gene approaches use gene therapy to alter the function of mature cells. Commonly used somatic gene therapy strategies include the overexpression of an existing gene and/or the insertion of smaller nucleic acid sequences into cells to alter the production of an existing gene. ALI/ARDS may be suitable for gene-based therapies as it is an acute but relatively transient process [8] , requiring short-lived gene expression, obviating the need for repeated therapies and reducing the risk of an adverse immunological response. Th e distal lung epithelium is selectively accessible via the tracheal route of administration, allowing targeting of the pulmonary epithelium [9] . Th e pulmonary vasculature is also relatively accessible, as the entire cardiac output must transit this circulation. Antibodies that bind antigens selectively expressed on the pulmonary endothelial surface can be complexed to gene vectors to facilitate selective targeting following intravenous administration [10] . It is also possible to use gene-based strategies to target other cells central to the pathogenesis of ALI/ARDS, such as leuko cytes and Abstract Acute lung injury (ALI) and acute respiratory distress syndrome (ARDS) confer substantial morbidity and mortality, and have no specifi c therapy. The accessibility of the distal lung epithelium via the airway route, and the relatively transient nature of ALI/ ARDS, suggest that the disease may be amenable to gene-based therapies. Ongoing advances in our understanding of the pathophysiology of ALI/ARDS have revealed multiple therapeutic targets for genebased approaches. Strategies to enhance or restore lung epithelial and/or endothelial cell function, to strengthen lung defense mechanisms against injury, to speed clearance of infection and to enhance the repair process following ALI/ARDS have all demonstrated promise in preclinical models. Despite three decades of gene therapy research, however, the clinical potential for gene-based approaches to lung diseases including ALI/ ARDS remains to be realized. Multiple barriers to eff ective pulmonary gene therapy exist, including the pulmonary architecture, pulmonary defense mechanisms against inhaled particles, the immunogenicity of viral vectors and the poor transfection effi ciency of nonviral delivery methods. Defi cits remain in our knowledge regarding the optimal molecular targets for genebased approaches. Encouragingly, recent progress in overcoming these barriers off ers hope for the successful translation of gene-based approaches for ALI/ARDS to the clinical setting. fi bro blasts [11] . Furthermore, gene-therapy-based approaches off er the potential to selectively target diff erent phases of the injury and repair process. Th e potential to target specifi c aspects of the injury and repair processes such as epithelial-mesenchymal transition, fi brosis, fi brinolysis, coagulopathy and oxidative stress with these approaches is also clear. Gene therapy requires the delivery of genes or smaller nucleic acid sequences into the cell nucleus using a carrier or vector. Th e vector enables the gene to overcome barriers to entry into the cell, and to make its way to the nucleus to be transcribed and translated itself or to modulate transcription and/or translation of other genes. Both viral and nonviral vector systems have been developed (Table 1) . Viral vectors are the most eff ective and effi cient way of getting larger nucleic acid sequences, particularly genes, into cells (Table 1) . Th e viral genome is modifi ed to remove the parts necessary for viral replication. Th is segment is then replaced with the gene of interesttermed a transgene -coupled to a promoter that drives its expression. Th e modifi ed genome is then encapsulated with viral proteins. Following delivery to the target site, the virus binds to the host cell, enters the cytoplasm and releases its payload into the nucleus (Figure 1 ). Th e size of trans gene that can be used depends on the capsid size. A number of diff erent viral vectors have been used in preclinical lung injury studies to date. Adenoviruses have double-stranded DNA genomes, have demonstrated promise in preclinical models [12, 13] and are well tolerated at low to intermediate doses in humans [14, 15] . Advantages include their ease of production, the high effi ciency at which they can infect the pulmonary epithelium [14, 16] and that they can deliver relatively large transgenes. A disadvantage of adenoviruses is their immunogenicity, particularly in repeated doses [14] . Newer adenoviral vectors, in which much of the immuno genicity has been removed, hold promise [17] . While adenovirus-mediated gene transfer in the absence of epithelial damage is relatively ineffi cient [18] , this may be less of a problem in ALI/ARDS that is characterized by widespread epithelial damage. Adeno-associated viruses (AAVs) are single-stranded DNA parvoviruses that are replication defi cient [19] . A substantial proportion of the human population has been exposed to AAVs but the clinical eff ects are unknown. AAV vectors have a good safety profi le, and are less immunogenic compared with other viruses, although anti bodies do develop against AAV capsid proteins that can compromise repeat administration. AAV vectors can insert genes at a specifi c site on chromosome 19 . Th e packaging capacity of the virus is limited to 4.7 kb, restricting the size of the transgene that can be used. AAVs are less effi cient in transducing cells than adenoviral vectors. Successful AAV vector gene transfer has been demon strated in multiple lung cell types including lung progenitor cells, in both normal and naphthaleneinduced ALI lungs [20] . AAV serotypes have specifi c tissue tropisms, due to diff erent capsid proteins that bind to specifi c cell membrane receptors. AAV-5 [21] and AVV-6 [22] exhibit enhanced tropism for the pulmonary epi thelium [21, 22] . AAVs can transduce nondividing cells and result in long-lived transgene expression. AAV vectors have been used in clinical trials in cystic fi brosis patients, underlining their safety profi le [23, 24] . Th ese RNA viruses can transfect nondividing cells such as mature airway epithelial cells [25] . Th e virus stably but randomly integrates into the genome and expression is likely to last for the lifetime of the cell (~100 days). Th e transgene can be transmitted post mitosis, and there is also a risk of tumorigenesis if the transgene integrates near an oncogene. Th e development of leukemias in children following gene therapy for severe combined immunodefi ciency highlights this risk [26, 27] . While lentiviral vectors may be useful to correct a gene defi ciency associated with increased risk of ALI, the long-lived gene expression of lentiviral delivered genes may be more suitable for chronic diseases than for ALI/ARDS. Nonviral delivery systems, while generally less effi cient than viral vectors in transfecting the lung epithelium, are increasingly used to deliver smaller DNA/RNA molecules (Table 1 ). Strategies include the use of DNA-lipid and DNA-polymer complexes and naked DNA/RNA oligonucleotides, such as siRNA [28] , decoy oligo nucleo tides [29] and plasmid DNA [30] . Nonviral delivery systems are less immunogenic than viral vector-based approaches, and can be generated in large amounts at relatively low cost. Plasmid vectors are composed of closed circles of doublestranded DNA. As naked and plasmid DNA contain no proteins for attachment to cellular receptors, there is no specifi c targeting to diff erent cell types and thus it is essential that the DNA is placed in close contact with the desired cell type. Th ese limitations make this approach less relevant clinically. Th e therapeutic DNA is held within a sphere of lipids, termed a lipoplex, or within a sphere of polymers, such as polyethyleneimine, termed a polyplex. Lipoplexes and polyplexes act to protect the DNA, facilitate binding to the target cell membrane and also trigger endocytosis of the complex into the cell, thereby enhancing gene expression. Th ese systems can be modifi ed to include a targeting peptide for a specifi c cell type, such as airway epithelial cells [31] . Th ese complexes effi ciently and safely transfect airway epithelial cells [31] , and they have demonstrated promise in human studies [32] . siRNAs are dsRNA molecules of 20 to 25 nucleotides that can regulate the expression of specifi c genes. Specifi c siRNAs reduce infl ammation-associated lung injury in Table 1 . Viral vector-delivered gene therapy Relatively easily produced Immunogenic [14] Adenoviral transfer of genes for a surfactant (dsDNA genome) Effi ciently transfect lung enzyme [49] , angiopoietin-1 [51] , HSP-70 [52] , epithelium [14, 16] apolipoprotein A-1 [53] , and Na + ,K + -ATPase pump Can deliver larger genes [55] genes attenuate experimental ALI Well tolerated in lower doses [1, 3] Adenoviral delivery of IL-10 gene attenuates zymosan ALI at low doses, but is harmful at high doses [58] Adeno-associated virus Good safety profi le; less Limited transgene size AAV vector gene transfer demonstrated in multiple vectors (ssDNA genome) immunogenic Diffi cult to produce in large lung cell types including progenitor cells in both Inherently replication defi cient quantities normal lungs and following naphthalene-induced AAV-5 and AAV-6 lung epithelial ALI [20] tropism [10, 11] Transduce nondividing cells Long-lived gene expression Used in clinical trials for CF [12, 13] Lentivirus vectors Transduce nondividing cells [25] Oncogenesis risk due to Lentiviral transfer of shRNA to silence CD36 gene (RNA genome) Integrate stably but randomly integration into genome expression suppresses silica-induced lung fi brosis into the genome [26, 27] in the rat [35] Nonviral gene-based strategies Plasmid transfer (closed Easily produced at low cost No specifi c cell targeting Electroporation-mediated gene transfer of the dsDNA circles) Very ineffi cient Na + ,K + -ATPase rescues endotoxin-induced lung injury [60] Nonviral DNA complexes Complexes protect DNA Less effi cient than viral vectors Cationic lipid-mediated transfer of the Na + ,K + -(lipoplexes or polyplexes) Complexes facilitate cellular ATPase gene ameliorated high-permeability targeting [31] pulmonary edema [59] Lipoplex-delivered IL-10 gene decreased CLP-induced ALI [61] Systemic cationic polyethylenimine polyplexes incorporating indoleamine-2,3-dioxygenase decreased ischemia-reperfusion ALI [62] DNA and RNA Easily produced at low cost No specifi c cell targeting Specifi c siRNAs reduce infl ammation-associated oligonucleotides (siRNA, Smaller molecules that can lung injury in humans [33] and in animal models shRNA, decoy easily enter cells [28, 34] oligonucleotides) Target regulation of specifi c genes shRNA-based approaches have reduced lung injury in animal models [35, 36] Cell-delivered gene therapy humans [33] and in animal models [28, 34] . shRNA is a single strand of RNA that, when introduced into the cell, is reverse transcribed and integrated into the genome, becoming heritable. During subsequent transcription, the sequence generates an oligonucleotide with a tight hairpin turn that is processed into siRNA. shRNAs have reduced lung injury in animal models [35, 36] . Decoy oligonucleotides are double-stranded DNA molecules of 20 to 28 nucleo tides, which bind to specifi c transcription factors to reduce expression of targeted genes, and have been successfully used in animal models [37, 38] . An alternative approach is to use systemically delivered cells to deliver genes to the lung. Th is approach has been used to enhance the therapeutic potential of stem cellssuch as mesenchymal stem/stromal cells, which demon strate promise in preclinical ALI/ARDS models [39] . Fibroblasts have also been used to successfully deliver genes to the lung to attenuate ALI [40] . Preliminary data from a clinical trial in pulmonary hypertension show that endothelial progenitor cells overexpressing endothelial nitric oxide synthase (NOS3) decrease pulmonary vascular resistance [41] , highlighting the potential of cell-delivered gene therapy for ALI/ARDS. Nebulization of genetic material into the lung is eff ective [42] , safe and well tolerated [32, 43, 44] . Th e integrity of AAV vectors [9, 43] and adenoviral virus vectors [44] are maintained post nebulization, as are cationic lipid vectors [32] and DNA and RNA oligonucleotides [45] . A number of gene therapy clinical trials have utilized nebulization to deliver the transgene to the lung [23, 43] , but without clear clinical benefi t to date [43, 44] . Intravascular delivery approaches target the lung endothelium. Th ese approaches have been successfully used in preclinical studies of cell-based gene therapies [39, 40] , and also with vectors that incorporate components such as antibodies to target antigens on the lung endothelium [10] . Successful gene-based therapies require the delivery of high quantities of the gene or oligonucleotide to the pulmonary epithelial or endothelial surface, require effi cient entry into the cytoplasm of these large and insoluble nucleic acids, which then have to move from the cytoplasm into the nucleus, and activate transcription of its product. Multiple barriers exist that hinder this process, not least the natural defense mechanisms of the lung, and additional diffi culties that exist in transducing the acutely injured lung (Table 2 ). Limitations regarding delivery technologies and defi ciencies in our knowledge regarding the optimal molecular targets also reduce the effi cacy of these approaches. Th e lung has evolved eff ective barriers to prevent the uptake of any inhaled foreign particles [46] . While advantageous in minimizing the potential for uptake of external genetic material (for example, viral DNA), these barriers make it more diffi cult to use gene-based therapies in the lung. Barriers to entry of foreign genetic material into the lung include airway mucus and the epithelial lining fl uid, which traps and clears inhaled material. Th e glycocalyceal barrier hinders contact with the cell membrane, while the tight intercellular epithelial junctions and limited luminal endocytosis further restrict entry of foreign material into the epithelial cells. Transducing the acutely injured lung may be diffi cult, due to the presence of pulmonary edema, consolidated or collapsed alveoli, and additional extracellular barriers such as mucus. Gene-based therapies targeted at the pulmonary epithelium may be less eff ective where there is extensive denudation of the pulmonary epithelium, as may occur in primary ARDS. Encouragingly, there is some evidence to suggest that ALI may not substantially impair viral gene transfer to the alveolar epithelium [47] . Th e key limitation of nonviral vector approaches has been their lack of effi ciency in mediating gene transfer and transgene expression in the airway epithelium. Viral vectors are immunogenic, due to the protein coat of the viral vector, and the immune response is related to both vector dose and number of administrations. Th e potential to limit administration to a single dose in ALI/ARDS may reduce this risk. However, the development of an infl amma tory response resulting in death following administration of a fi rst-generation adenoviral vector highlights the risks involved [48] . Additional limitations of viral vectors include transgene size, which is limited by the size of the capsid that encloses the viral genes. Th e therapeutic potential of gene therapy for ALI/ARDS is underlined by a growing body of literature demon strating effi cacy in relevant preclinical models. In considering the clinical implications of these studies, it is important to acknowledge that animal models of ARDS do not fully replicate the complex pathophysiological changes seen in the clinical setting. Th is is highlighted by the fact that many pharmacologic strategies demonstrating considerable promise in preclinical studies were later proven ineff ective in clinical trials. Nevertheless, these studies provide insights into the clinical potential of these strategies. Adenovirus-mediated transfer of a gene that enhances surfactant production improves lung function and confers resistance to Pseudomonas aeruginosa infection ( Figure 2 ) [49] . Adenovirus-delivered superoxide dismutase and catalase genes protected against hyperoxic-induced, but not ischemia-reperfusion-induced, lung injury [50] . More recent studies have demonstrated the therapeutic potential of overexpression of a number of genes, including angio poietin-1 [51] , HSP-70 [52] , apolipo protein A-1 [53] , defensin β2 [54] and the Na + ,K + -ATPase pump [55] . In contrast, overexpression of IL-1β can directly cause ALI [56] , while overexpression of suppressor of cytokine signal ing-3 worsens immune-complex-induced ALI [57] . Intriguingly, intra tracheal administration of adenoviral vector incor porating IL-10, prior to zymosan-induced lung injury, improved survival at a lower dose but was ineff ective and even harmful at higher doses [58] . An early murine study demonstrated that cationic lipidmediated transfer of the Na + ,K + -ATPase gene ameliorated high-permeability pulmonary edema [59] . Electroporationassisted gene transfer of plasmids encoding for Na + ,K + -ATPase reverses endotoxin-induced lung injury [60] . Th e lipoplex-delivered IL-10 gene decreased lung and systemic organ injury induced by cecal ligation and puncture in mice [61] . Systemically administered cationic polyethyleni mine polyplexes incorporating indoleamine-2,3-dioxyge nase transduced pulmonary endo thelial cells and decreased lung ischemia-reper fusion injury [62] . NF-κB decoy oligonucleotides, incorporated into viral vectors, attenuate systemic sepsis-induced lung injury when administered intravenously (Figure 3 ) [37] . In animal models, both intratracheal [34, 63] and intra venously [29, 64] administered siRNA successfully silence their target genes. shRNA-based approaches have been used to suppress silica-induced lung fi brosis [35] and to ameliorate lung ischemia-reperfusion-induced lung injury [36] . More recently, aerosolization of siRNA that targets respiratory syncytial virus viral replication was safe and potentially eff ective in patients post lung transplant with respiratory syncytial virus infection [33] , clearly illustrating the therapeutic potential of these approaches for ALI/ARDS. Mei and colleagues enhanced the effi cacy of mesen chymal stem/stromal cells in endotoxin-induced ALI by transducing them to overexpress angiopoeitin-1 (Figure 4 ) [39] . Mesenchymal stem/stromal cells overexpressing IL-10 decreased alveolar infi ltration of CD4 and CD8 T cells following lung ischemia-reperfusion injury [65] . Bone marrow stem cells expressing keratinocyte growth factor attenuate bleomycin-induced lung injury [66] . Non stem cells can also be used to deliver genes to the injured lung [67] . Fibroblasts overexpressing angiopoeitin-1 attenuate endotoxin-induced lung injury [40] , while fi broblasts overexpressing vascular endothelial growth factor and endothelial nitric oxide synthase can attenuate or even reverse endotoxin-induced ALI [68] . Advances in the identifi cation of therapeutic targets, improvements in viral and nonviral vector technologies, and regulation of gene-based therapies by temporal and spatial targeting off er the potential to translate the therapeutic promise of gene-based therapies for ALI/ ARDS to the clinical setting (Table 3) . Viral vectors remain the focus of intensive research to optimize their effi ciency, to minimize their immuno genicity and to enhance their tissue specifi city [19, 31, 69, 70] . Strategies to develop less immunogenic vectors have focused on modifying the naturally occurring proteins in the viral coat [71] . Much research has been devoted to searching and characterizing both naturally occurring [71] and engineered capsid variants from mammalian species [72] . Capsid protein modification has also been used to enhance tissue specifi city [70] . Envelope protein pseudotyping involves encapsulating the modifi ed genome from one virus, such as simian immuno defi ci ency virus, with envelope proteins from another virus, such as vesicular stomatitic virus. Th is encapsu lation can enhance the therapeutic potential of viral vectors, by combining the advantages of one viral genome (for example, bigger payload or site-specifi c integration) with the tissue tropism of another virus. Strategies to enhance the eff ectiveness of the lipoplexes used to deliver plasmids and other DNA/RNA oligonucleotides involve manipulation of the lipoplex lipid content and the use of targeting peptides. Th e choice of lipid infl uences expression effi ciency by enhancing release of the genetic material within the target cell [73, 74] . Targeting peptides increases transfection effi ciency by directing the lipid to a particular cell membrane or cell type [31] . Physical methods of plasmid delivery such as electroporation [60] and ultrasound can enhance gene transfer by bringing the plasmid DNA into closer proximity with the cell membrane and/or causing temporary disruption of the cell membrane. Other physical methods can also be used to increase in vivo gene transfer, including pressurized vascular delivery, laser, magnetic fi elds and gene gun delivery. Th ese systems enable plasmid-based gene delivery to reach effi ciencies close to that achieved with viral vectors. Successful gene therapy relies upon being able to target the injury site, and to control the duration and levels of gene expression. Modifying the transgene DNA to exclude nonmethylated CpG motifs, typical of bacterial DNA, decreases the immune response and may increase transgene expression [75, 76] . High-effi ciency tissue-specifi c promoters may improve the effi ciency and specifi city of transgene expression. Lung-specifi c promoters include surfactant promoters [77] such as the surfactant protein C promoter [78] , a ciliated cell-specifi c promoter FOXJ1 [79] , the cytokeratin 18 promoter [80] , and the Clara cell 10-kDa protein [78] . Promoters can also be used to target a specifi c phase of illness, switching on when required to produce an eff ect at the optimal time point. A related approach is the development of promoters that allow for transfected genes to be turned on and off . Currently, the tetracycline-dependent gene expression vector [81] is the most widely used regulated system as it has a good safety profi le. Tetracycline is rapidly metabolized and cleared from the body, making it an ideal drug to control gene expression. However, the potential for an activator such as tetracycline to modulate the lung injury should be borne in mind. New-generation transactivators, with no basal activity and increased sensitivity, have now been developed [82] . In an ARDS context, conditional regulation of gene expression by the combined use of a lung-specifi c promoter and the tetracycline-dependent gene expression system may be a useful approach [83] . Capsid protein modifi cation to reduce immunogenicity [71] Capsid protein modifi cation to enhance tissue specifi city [70] Envelope protein pseudotyping Manipulation of lipoplex lipid content to enhance cellular uptake [73, 74] Use of targeting peptides on lipoplexes and polyplexes [31] Strategies to enhance gene transfer; for example, electroporation, ultrasound, gene gun delivery Modifying transgene DNA to eliminate bacterial motifs [75, 76] Development of high-effi ciency tissue-specifi c promoters [77] [78] [79] [80] Development of promoters that regulate gene expression [83] Enhanced therapeutic targeting Nebulization technologies [9] Strategies to target the pulmonary endothelium [10] Improved cellular uptake of vector Surface active agents to enhance vector spread [84] Reduce ubiquitination of viral capsid proteins [85] Better therapeutic targets Enhancement or restoration of lung epithelial and/or endothelial cell function [86] Strengthening lung defense mechanisms against injury [87] Speeding clearance of infl ammation and infection Enhancement of the repair process following ALI/ARDS [88] . An advantage of gene-based strategies is the ability to target specifi c cells within an organ; for example, the epithelial cells of the lung. Novel nebulization technologies, which facilitate the delivery of large quantities of undamaged vector to the distal lung, demonstrate considerable promise in this regard [9] . Alternative approaches to spatial targeting include targeting specifi c receptors that are plentiful on the target cell to increase transfection effi ciency. An interesting development in this regard is the targeting of systemically administered therapies to the pulmonary endothelium using antibodies to proteins expressed preferentially on these cells ( Figure 5 ) [10] . In these studies, the antioxidant enzyme catalase was conjugated with antibodies to the adhesion molecule PECAM, which is widely expressed on pulmonary endothelial cells, and to a nonspecifi c IgG antibody. Th e anti-PECAM/catalase conjugate, but not the IgG/catalase conjugate, bound specifi cally to the pulmonary endothelium and attenuated hydrogen peroxide injury. Specifi c strategies have been developed to maximize uptake of vector into alveolar epithelial cells. It is possible to enhance lung transgene expression with the use of surface-active agents such as perfl urocarbon, which enhances the spread of vector and mixing within the epithelial lining fl uid [84] . Agents that reduce ubiquitination of AAV capsid proteins following endocytosis, such as tripeptide proteasome inhibitors, dramatically augment (>2,000-fold) AAV vector transduction in airway epithelia [85] . Ultimately, the success or failure of gene-based therapies for ALI/ARDS is likely to rest on the identifi cation of better gene targets. Ongoing advances in our understanding of the pathophysiology of ALI/ARDS continue to reveal novel therapeutic targets for gene-based approaches. Promising potential approaches include strate gies to enhance or restore lung epithelial and/or endothelial cell function [86] , to strengthen lung defense mechanisms against injury [87] , to speed clear ance of infl ammation and infection, and to enhance the repair process following ALI/ARDS [88] . ALI/ARDS may be a particularly suitable disease process for gene-based therapies (Table 4 ). Th is is supported by increasing evidence from relevant preclinical ARDS models for the effi cacy of gene-based therapies that enhance or restore lung epithelial and/or endothelial cell function, strengthen lung defense mecha nisms against injury, speed resolution of infl ammation and infection, and enhance the repair process following ALI/ARDS. Despite this promising preclinical evidence, the potential for gene based approaches to ALI/ARDS in the clinical setting remains to be realized. Multiple barriers exist to the successful use of gene-based therapies in the lung, which limit the effi cacy of these approaches. Future research approaches should focus on overcoming these barriers, by developing more eff ective and less immunogenic vector delivery systems, developing strategies to focus gene expression on specifi c injury zones of the lung for defi ned time periods, and identifying better molecular targets that can take advantage of these potentially very powerful therapeutic approaches. Abbreviations AAV, adeno-associated virus; ALI, acute lung injury; ARDS, acute respiratory distress syndrome; IL, interleukin; NF, nuclear factor; shRNA, small hairpin RNA; siRNA, small interfering RNA. The authors declare that they have no competing interests.

Critical care services and the H1N1 (2009) influenza epidemic in Australia and New Zealand in 2010: the impact of the second winter epidemic

INTRODUCTION: During the first winter of exposure, the H1N1 2009 influenza virus placed considerable strain on intensive care unit (ICU) services in Australia and New Zealand (ANZ). We assessed the impact of the H1N1 2009 influenza virus on ICU services during the second (2010) winter, following the implementation of vaccination. METHODS: A prospective, cohort study was conducted in all ANZ ICUs during the southern hemisphere winter of 2010. Data on demographic and clinical characteristics, including vaccination status and outcomes, were collected. The characteristics of patients admitted during the 2010 and 2009 seasons were compared. RESULTS: From 1 June to 15 October 2010, there were 315 patients with confirmed influenza A, of whom 283 patients (90%) had H1N1 2009 (10.6 cases per million inhabitants; 95% confidence interval (CI), 9.4 to 11.9) which was an observed incidence of 33% of that in 2009 (P < 0.001). The maximum daily ICU occupancy was 2.4 beds (95% CI, 1.8 to 3) per million inhabitants in 2010 compared with 7.5 (95% CI, 6.5 to 8.6) in 2009, (P < 0.001). The onset of the epidemic in 2010 was delayed by five weeks compared with 2009. The clinical characteristics were similar in 2010 and 2009 with no difference in the age distribution, proportion of patients treated with mechanical ventilation, duration of ICU admission, or hospital mortality. Unlike 2009 the incidence of critical illness was significantly greater in New Zealand (18.8 cases per million inhabitants compared with 9 in Australia, P < 0.001). Of 170 patients with known vaccination status, 26 (15.3%) had been vaccinated against H1N1 2009. CONCLUSIONS: During the 2010 ANZ winter, the impact of H1N1 2009 on ICU services was still appreciable in Australia and substantial in New Zealand. Vaccination failure occurred.

Influenza A H1N1 2009 emerged in Mexico in early 2009 and spread rapidly causing a pandemic. The World Health Organization (WHO) declared a phase 6 influenza pandemic on 11 June 2009 and declared it to be over on 10 August 2010 [1] . The first wave of the H1N1 2009 outbreak was notable for the number of fatal cases among young people and atypical risk factors for developing severe diseases. Since 19 April 2009, WHO had reported over 491,766 laboratory-confirmed cases of H1N1 2009 and 18,449 related deaths [1] . People in Australia and New Zealand (ANZ) were significantly affected by the virus, with a total of 43,700 confirmed cases in Australia as of October 2010, with 6,064 cases occurring between 1 January and 15 October 2010 [2] . We have reported previously the serious impact of the virus on the provision of critical care services in 2009 [3] . In order to rapidly inform health professionals in the Northern Hemisphere this study censored new incident cases before the end of the influenza season. In 2010, the deployment of vaccination and the acquisition of natural immunity against H1N1 2009 were expected to decrease the burden of disease due to influenza [4] . Australia and New Zealand were among the first countries to experience a second influenza season with H1N1 2009 following widespread deployment of vaccination. In this report, we describe the incidence of intensive care unit (ICU) admissions, demographic and clinical characteristics (including vaccination status) and outcome of all patients with laboratory confirmed H1N1 2009 admitted to ICUs in ANZ during the second winter (2010) of this influenza epidemic. We compare these characteristics with those of patients admitted to ICU during the corresponding period of 2009. We performed a multicentre study in 187 ICUs in ANZ comprising all adult, paediatric and combined adult and paediatric ICUs [5] . These ICUs had a total of 1,821 beds, of which 1,487 were equipped for mechanical ventilation. Each centre or region obtained Ethics Committee approval and the requirement for individual subject informed consent was waived at all sites. We report our findings according to STROBE guidelines for observational studies [6] . Between 1 June and 15 October, in both 2009 and 2010, we screened for patients admitted to ICU with confirmed influenza A. Influenza A was confirmed by reverse transcriptase polymerase chain reaction (RT-PCR), antigen detection, or serology. H1N1 2009 and seasonal subtypes, H1N1 and H3N2, were determined by RT-PCR or specific serology. The laboratories were accredited by the National Association of Testing Authorities in Australia or by International Accreditation New Zealand. Population data for Australia and New Zealand were obtained from Australian Bureau of Statistics [7] and Statistics New Zealand [8] for 2009 and 2010. We collected patient-specific data as described previously [3] , although vaccination status against H1N1 2009 virus was collected only during 2010. We divided patients into the age groups used in a previous report [9] . We calculated the duration of ICU and hospital stay and ICU occupancy rates for Australia and New Zealand. We recorded patient outcomes at ICU and hospital discharge status or as still in hospital or in ICU as of 15 October 2010 for patients admitted in 2010 and as of 23 November 2009 for patients admitted in 2009. Finally, we obtained data on vaccination in both countries. We collected data using electronic case report forms. The study-coordinating centre was the Australian and New Zealand Intensive Care-Research Centre, Monash University, Melbourne, Australia [10] . H1N1 2009 infection is subject to mandatory reporting in both Australia and New Zealand and wherever possible diagnoses were confirmed with the relevant public health authorities. In addition, to confirm the completeness of case ascertainment, we contacted all ICUs that had no reported cases at the end of each study period. Cases transferred between ICUs were counted as a single ICU case. We made no assumptions for missing data and all proportions were calculated as percentages of available data. We performed statistical analysis using SAS version 9.1 (SAS Institute Inc., Cary, NC, USA). We calculated descriptive statistics for all study variables. We report continuous variables as medians with interquartile range (IQR) and categorical variables as percentages with 95% confidence interval (95% CI) where appropriate. We estimated age-based population admission rates [7, 11] . We compared binomial variables of the second winter (2010) with those of the first winter (2009) using Chisquare tests for equal proportion or Fisher's Exact test where numbers were small. Comparisons between continuous variables were made using Wilcoxon rank sum test. A two-sided P-value of < 0.05 was considered to be statistically significant. During the study period, from 1 June until 15 October 2010, 315 patients with confirmed influenza A were admitted to an ICU, compared with 1,113 patients during the same corresponding period in 2009. The predominant sub-type was H1N1 2009 in both years (90% in 2010 versus 83% in 2009, P = 0.002). The distribution of sub-types of influenza A is reported in Table 1 . In 2010 there were 283 patients with confirmed H1N1 2009 admitted to ICU, corresponding to a population incidence of admission to ICU of 10.6 (95% CI, 9.4 to 11.9) per million inhabitants [7, 8] . By comparison, during 2009 there were almost three times as many patients with confirmed H1N1 2009 (n = 921), corresponding to a population incidence of admission to ICU of 35.2 (95% CI, 32.9 to 37.5) per million inhabitants [7, 8] (P < 0.001). The geographical distribution of admissions was also different between 2010 and 2009 (Table 1 ). In 2010 the population incidence in Australia was 2-fold lower than in New Zealand (P < 0.001), whereas in 2009 this relationship was reversed with a 1.3-fold higher incidence in Australia (P = 0.009). The overall incidence, combining 2009 and 2010, was 45.3 (95% CI, 42.4 to 48.2) admission to ICU per million inhabitants in Australia and 47.1 (95% CI, 40.5 to 53.7) in New Zealand (P = 0.65). The number of patients admitted to an ICU with H1N1 2009 according to study week is displayed in Figure 1 . In 2010 the onset of the epidemic was delayed (by five weeks) and the peak incidence was lower. The impact on ICU services was significantly lower in 2010 with peak daily ICU bed occupancy being 7.5 (95% CI, 6.5 to 8.6) per million inhabitants in 2009 compared with 2.4 (95% CI, 1.8 to 3.0) per million inhabitants in 2010 (P < 0.001) ( Table 1) . The clinical characteristics, risk factors, and outcomes of patients admitted to ICUs with H1N1 2009 were broadly similar in 2010 to those admitted in 2009 ( Table 2 ). Comparing 2010 with 2009 there were no significant differences in the distribution of cases among different age groups. In both study periods the highest number of ICU admissions occurred among patients aged between 25 and 49. There was no difference in the proportion of patients with a body mass index greater than 35 kg/m 2 , who were pregnant or post-partum or who had no known predisposing factor between that observed in 2009 and 2010 ( Table 2) . The proportion of patients with asthma or chronic obstructive pulmonary disease, chronic heart failure, or an Acute Physiology Age Chronic Health Evaluation (APACHE) III or paediatric co-existing illness was lower in 2010, but these factors were still over-represented in comparison to the general population, as they were in 2009 [3] (Table 2) admissions due to confirmed influenza in both countries [3] . The second wave occurred later in the year but resulted in similar illness severity, affected similar groups at risk and caused similar in-hospital mortality. The incidence of H1N1 2009 influenza critical illness during 2010 was higher in NZ than in Australia, which was a reversal of the pattern in 2009. However, the combined incidence over both winters was similar in both countries. We observed vaccination failure in a substantial proportion of patients for whom vaccination status was known. Our observations do not support the view that the H1N1 2009 pandemic has come to an end [13] . Although the incidence of critical illness was significantly lower in 2010 the pattern of illness among patients who were admitted to an ICU was similar to that observed in 2009 and in other reports of patients admitted to ICUs [14] [15] [16] . ARDS was present in more than 50% of patients, as previously described in ANZ [3] and worldwide [14, 15, [17] [18] [19] [20] [21] . The length of stay in ICU was unchanged [3] and similar to that reported elsewhere [14, 20] . The risk factors for admission to ICU were similar to those reported during the first wave [3, 22, 23] . In addition, the treatments administered and the mortality that occurred were similar to that observed in 2009 in ANZ as well as elsewhere [3, 13, 14, 18] . The incidence of ICU admission per million inhabitants was higher in Australia than in New Zealand in 2009 (36.5 versus 28.3 admissions to ICU per million inhabitants) [3] . Conversely, in 2010 this incidence was higher in New Zealand. This difference is concordant with national data reporting 6,064 confirmed H1N1 cases in Australia (272 cases per million inhabitants) versus 1,810 in New Zealand (415 cases per million inhabitants in 2010) during the same period [2, 24] . The proportion of confirmed cases hospitalised (727 of 1,810 = 40.1%) [25] , and of hospitalised cases admitted to ICU (82 of 732 = 11.2%) in New Zealand was similar to 2009 [26] . Accordingly, the higher ICU admission rate per population observed in 2010 was due to a higher incidence of community infection, rather than a difference in severity of disease. The reason for the difference in infection rate between Australia and New Zealand in 2010 is unclear, as a similar proportion of the population was seropositive after the 2009 pandemic wave in Australia (22% (95% CI 19.1 to 24.9)) compared to New Zealand (26.7% (95% CI 22.6 to 29.4)) [27, 28] and a similar proportion were vaccinated (21.8% versus 24.1%, respectively) during the inter-wave period. Possible explanations include natural variations in community spread of influenza or an effect of the difference in vaccination deployment where Australia used both early monovalent and later polyvalent vaccines while New Zealand relied predominantly on delivery of the polyvalent vaccine alone, although overall coverage was similar. We found that 15.3% of those patients for whom data were available had been previously vaccinated, which is consistent with another report of H1N1 2009 vaccination effectiveness [29] . In a critically ill patient with a severe pneumonia and a history of H1N1 vaccination the possibility of vaccination failure and H1N1 2009 infection should be considered. Our data are subject to some limitations. To make this report available in a timely manner, we censored hospital outcome data. Ascertainment of cases of H1N1 2009 admitted to ICUs in 2010 as well as in 2009 may not have been complete, and we cannot exclude the possibility that a small number of cases were not reported to the registry, and false negative diagnostic tests may well have underestimated the true burden of H1N1 2009 in our patients. Among the patients with confirmed influenza A, there were 29 in whom the influenza was not sub-typed in 2010 and 132 in 2009. Thresholds for undertaking testing both in hospital and in the community were not standardised. It is not possible to reach a conclusion regarding the efficacy of vaccination as the vaccination status of many patients was unknown. Finally, while we report a similar severity of the illness to 2009, we do not have information about anti-viral treatment, potential viral mutation and resistance to anti-viral drugs. Finally, we did not evaluate the role played by a corticosteroid therapy on the outcome of patients with ARDS, and we were not able to cope with the controversy about the effect of this treatment [30] . In conclusion, the impact of the second H1N1 2009 winter epidemic was still substantial although significantly less than in 2009. It had a lower peak occurring approximately five weeks later than in 2009, affected similar individuals, was similar in clinical severity, carried a similar mortality rate. In patients with H1N1 2009 infection requiring ICU admission a number of apparent vaccination failures were observed. Based on these data and despite the deployment of the vaccination, a second season of H1N1 2009 influenza may still have a substantial intensive care impact. • H1N1 (2009) had a substantial impact on ICU resources during the winter of 2010 in Australia and New Zealand. • Risk factors remain similar to those reported in 2009 and include obesity, pregnancy and presence of comorbidity. • In 2010, the illness severity, reflected by treatment with mechanical ventilation, renal replacement therapy, vasopressor drug, and extracorporeal membrane oxygenation as well as by hospital mortality, was similar to that observed the previous year.

Multinational, observational study of procalcitonin in ICU patients with pneumonia requiring mechanical ventilation: a multicenter observational study

INTRODUCTION: The intent of this study was to determine whether serum procalcitonin (PCT) levels are associated with prognosis, measured as organ dysfunctions and 28-day mortality, in patients with severe pneumonia. METHODS: This was a multicenter, observational study of critically ill adult patients with pneumonia requiring mechanical ventilation conducted in 10 academic hospitals in Canada, the United States, and Central Europe. PCT was measured daily for 14 days using an immuno-luminometric assay. RESULTS: We included 175 patients, 57 with community acquired pneumonia (CAP), 61 with ventilator associated pneumonia (VAP) and 57 with hospital acquired pneumonia (HAP). Initial PCT levels were higher in CAP than VAP patients (median (interquartile range: IQR); 2.4 (0.95 to 15.8) vs. 0.7 (0.3 to 2.15), ng/ml, P < 0.001) but not significantly different to HAP (2.2 (0.4 to 8.0) ng/ml). The 28-day ICU mortality rate for all patients was 18.3% with a median ICU length of stay of 16 days (range 1 to 142 days). PCT levels were higher in non-survivors than in survivors. Initial and maximum PCT levels correlated with maximum Sequential Organ Failure Assessment (SOFA) score r(2 )= 0.50 (95% CI: 0.38 to 0.61) and r(2 )= 0.57 (0.46 to 0.66), respectively. Receiver operating curve (ROC) analysis on discrimination of 28-day mortality showed areas under the curve (AUC) of 0.74, 0.70, and 0.69 for maximum PCT, initial PCT, and Acute Physiology and Chronic Health Evaluation (APACHE) II score, respectively. The optimal cut-off to predict mortality for initial PCT was 1.1 ng/ml (odds ratio: OD 7.0 (95% CI 2.6 to 25.2)) and that for maximum PCT was 7.8 ng/ml (odds ratio 5.7 (95% CI 2.5 to 13.1)). CONCLUSIONS: PCT is associated with the severity of illness in patients with severe pneumonia and appears to be a prognostic marker of morbidity and mortality comparable to the APACHE II score.

Respiratory tract infections requiring mechanical ventilation account for the majority of all infections treated in the intensive care unit (ICU) and are associated with prolonged hospital stay and high ICU mortality [1] [2] [3] . The Pneumonia Severity Index (PSI) is commonly used for risk stratification of patients with pneumonia. However, this parameter showed only moderate association with outcome prediction and was judged to be inadequate to guide clinical care [4] . Numerous studies have evaluated the diagnostic performance of invasive procedures, or of biochemical and molecular markers in blood or bronchoalveolar lavage (BAL) in patients with ventilator-associated pneumonia (VAP), hospital acquired pneumonia (HAP) and community acquired pneumonia (CAP). These methods are difficult to apply to daily clinical practice and none has proved to be predictive of outcome [5] [6] [7] [8] . Furthermore, many aspects in the strategies for diagnosing HAP and VAP especially regarding the importance of invasive procedures are still controversial [9, 10] . Indeed, a recent study revealed that use of invasive procedures for etiological diagnosis of pneumonia varies considerably between European ICUs [11] . This uncertainty is most likely responsible for antibiotic overtreatment observed in this group of patients [12, 13] . Thus, measures to aid the early identification of patients with pneumonia are underdeveloped. Such measures are needed as patients with pneumonia are at high risk of death and would benefit from early adaption of therapy. Procalcitonin (PCT), a relatively novel marker of infectious processes, has been shown to be associated with the severity of inflammation and prognosis during sepsis and septic shock [14] [15] [16] . In two large studies in the emergency department, low PCT-values were associated with a low risk of death in patients with CAP [17, 18] . Luyt and colleagues reported that PCT levels decreased during the clinical course of VAP but were significantly higher from Day 1 to Day 7 in patients with unfavorable outcomes [19] . The significance of PCT is emphasized by the observation that the course of PCT levels may safely guide antimicrobial therapy in patients with community acquired lower respiratory tract infections [20, 21] and ICU patients with suspected bacterial infections [22] . However, data about the significance of PCT in patients with hospital and ventilator acquired pneumonia requiring intensive care therapy are still limited. The aim of this multicenter study was to test the hypothesis that serum PCT levels can assist in identifying patients with severe pneumonia who are at increased risk of poor outcome, measured as organ dysfunction and 28-day mortality. In this multicenter, multi-national, observational study, patients admitted consecutively to the ICUs of 10 academic hospitals (8 in Canada and the United States and 2 in Europe) between 1 January 2003 and 20 November 2004 were screened for eligibility. The study protocol had been reviewed and approved by the Food and Drug Administration (protocol PCT-7; file number # I010023). Patients 18 years of age and older requiring mechanical ventilation with the new diagnosis of pneumonia within the last 48 hours were included. We excluded patients who were enrolled in a clinical study prior to baseline PCT sampling, had cardiogenic shock, had burns greater than 20% of total body surface, or were likely to die within 48 h, and postoperative patients following bone marrow transplant (within the last 6 months), coronary artery bypass grafts (within the last 7 days), and solid organ transplants (within the last 14 days). Patients were followed for 28 days after discharge from the ICU. The study was approved by local Institutional Review Boards/Ethics Committees of each participating institution and informed consent was obtained from the patients' next of kin. Pneumonia was defined as the presence of new or progressive infiltrate(s), consolidation, cavitation, or pleural effusion on chest radiographs and the new onset of at least two of the following signs or symptoms: 1) cough; 2) production of purulent sputum or a change in the character of sputum; 3) auscultatory findings on pulmonary exam of crackles and/or evidence of pulmonary consolidation (dullness on percussion, bronchial breath sounds); and/or 4) the presence of acute or progressive dyspnea, tachypnea, or hypoxemia. In addition, at least one of the following criteria had to be fulfilled to establish the diagnosis of pneumonia: 1) fever, defined as body temperature > 38°C (100.4°F) taken orally; > 38.5°C (101.2°F) tympanically; or > 39°C (102.2°F ) rectally or via pulmonary artery (PA) catheter; and/or 2) elevated total white blood count (WBC) > 10,000/ mm 3 , or > 15% immature neutrophils (bands), regardless of total WBC, or leukopenia with total WBC < 4,500/ mm 3 . Microscopic examination of the Gram stained respiratory secretions had to show the presence of microorganisms, with ≥25 polymorphonuclear cells and ≤10 squamous epithelial cells per field at 100× magnification (low-power, 10× objective). CAP [23] was defined as the occurrence of pneumonia in patients who had not resided in a long-term care facility for ≥14 days before the onset of symptoms and did not fulfill criteria of HAP, HAP [24] as pneumonia diagnosed in hospitalized patients or those residing in a long-term care facility (> 48 hours), such as a skilled nursing home facility or rehabilitation unit, or present < 7 days after a patient was discharged from the hospital with initial hospitalization of ≥3 days duration, and VAP [25] as pneumonia that developed more than 48 hours after intubation in mechanically ventilated patients who had no clinical evidence suggesting the presence or likely development of pneumonia at the time of intubation. Within 48 hours of enrolment, we sought to establish a diagnosis of pneumonia through culture and susceptibility testing of respiratory secretions obtained by deep expectoration, nasotracheal aspiration, intubation with endotracheal suctioning, bronchoscopy with BAL or protected-brush sampling, or transtracheal aspiration. The diagnosis could also be supported by culture of samples obtained by percutaneous lung or pleural fluid aspiration, and/or single diagnostic antibody titer, (IgM), or a four-fold increase in paired serum samples (IgG) for the presumed pathogen. Patients with burns greater than 20% of total body surface, expected death within 48 h, post bone marrow transplant within the last 6 months, cardiogenic shock, cardiovascular bypass within the last 7 days, solid organ transplant within the last 14 days, or patients participating in other studies were excluded. Key data were verified by source documents (hospital chart). Monitoring was conducted according to Good Clinical Practice (GCP) and standard operating procedures for compliance with applicable government regulations and was performed by an independent clinical research organization. We recorded demographic data including date of birth, gender, ethnic origin, weight, and height, type of pneumonia, and admission Acute Physiology and Chronic Health Evaluation (APACHE) II score at study enrolment. Organ dysfunction status was assessed daily as described elsewhere [26] and worst values of each calendar day were reported. A modified Sequential Organ Failure Assessment (SOFA) score that excluded the Glasgow Coma Scale (GCS) was utilized. PCT samples were collected for 14 days or until patients were discharged from the ICU and/or no longer required any mechanical ventilatory support. Blood samples not expected to be analyzed within 24 h of collection were frozen at -20°C for later analysis. PCT was measured using an immunoluminometric assay (LUMItest ® ; BRAHMS GmbH, Hennigsdorf, Germany). PCT levels were not available to the investigators until completion of the study and had no impact upon patient care during the course of the study. The primary objective was to detect a correlation between maximum PCT and SOFA-score. A total of 180 subjects were required in order to significantly demonstrate that the correlation coefficient is above 0.2 with a power of 90%. Means ± standard deviations (SDs) or medians with interquartile ranges (IQR) are reported as appropriate. The three types of pneumonia were compared using tie-corrected exact Kruskal-Wallis tests. Pair-wise comparisons of HAP and VAP to CAP were added, based on tie-corrected exact Mann-Whitney U-tests. Odds ratios and receiver operating characteristic (ROC) curve methodology were used to judge the predictive power of PCT for outcome. Of the 200 enrolled in this study, 25 patients were excluded from the analysis of the data. Of these, 21 patients had incomplete sampling and four patients met exclusion criteria. The characteristics on admission of the 175 patients included in our analysis study group are presented in Table 1 . Mean age was 62 years; roughly one-third had CAP, one-third had HAP, and one-third had VAP. The median hospital and ICU lengths of stay prior to enrolment were six days (range 0 to 368 days) and nine days (range 0 to 42 days), respectively. Patients with CAP had higher APACHE II and SOFA scores at inclusion than patients with VAP. Such a difference was not observed between VAP and HAP patients. The incidence of cardiovascular co-morbid conditions on admission to the ICU was lower in patients with VAP than in the other groups (Table 1) . Positive cultures of the microbiological samples taken within 48 h were reported in 119 patients (67.4%). Gram-positive organisms were isolated in 75 patients (42.9%) and Gram-negative organisms in 63 patients (36.0%). The detected microorganisms are shown in Time course of PCT levels PCT levels were elevated at the time of enrolment in all groups (Table 3) . Initial PCT levels were higher in CAP than VAP patients. The maximum PCT levels were higher in patients with CAP than those with HAP or VAP. Maximum PCT occurred a median of one to two days after inclusion into the study. As shown in Figure 1 , PCT levels were persistently higher in patients with CAP than those with HAP during the first week following inclusion. There was no difference of initial PCT levels in culture positive and culture negative patients ( The overall 28-day mortality rate was 18.3% (n = 32) and the median ICU length of stay (LOS) was 16 (9 to 28.5) days (range 1 to 142 days). The 28-day mortality was higher in patients with severe CAP compared with those with HAP or VAP (36.8% vs. 10.5% and 8.2%, respectively, P < 0.01 each). Likewise, the maximum degree of organ dysfunction as assessed by the maximum SOFA score was higher in CAP compared with HAP and VAP patients. PCT levels were consistently higher in non-survivors than survivors throughout the observation period ( Figure 2 ). Initial PCT values of VAP patients were significantly higher in non-survivors than in survivors with a median PCT of 0.6 ng/ml in the latter group (Figure 3) . This difference between survivors and non-survivors was also observed in HAP but did not reach statistical significance. In the survivors, PCT values dropped to a median of 50.0% (27.3 to 100.0%) of the baseline value (P < 0.001) during the first five study days. A drop of similar magnitude with 53.7% (27.6 to 148.0%) was observed in the non-survivors without reaching statistical significance (P = 0.08). Initial and maximal PCT levels correlated with maximum SOFA score (r 2 = 0. 51 and r 2 = 0.57, respectively). The association between initial and maximum PCT levels and SOFA score was independent of the type of pneumonia (Figure 4) . In a ROC analysis on discrimination of 28-day mortality, the area under the curves (AUC) for maximum PCT, initial PCT, and admission-day APACHE II score were 0.74, 0.70, and 0.69, respectively ( Figure 5 ). The AUCs were not statistically different. The best cut-off of initial PCT to predict 28-day mortality was 1.1 ng/ml (odds ratio 7.0 (95% CI 2.6 to 25.2)) and that of the maximum PCT was 7.8 ng/ ml (odds ratio 5.7 (95% CI 2.5 to 13.1)). The highest AUC was observed in VAP patients with 0.71 (95% CI 0.92 to 1.01) compared to CAP with 0.41 (95% CI 0.24 to 0.92) and HAP with 0.56 (95% CI 0.58 to 0.96). In this prospective multicenter study on a cohort of ICU-patients with severe pneumonia, median initial PCT levels were elevated above a normal value of 0.3 ng/ml in all groups. Those patients with ventilator associated pneumonia had the lowest initial PCT values. The maximum PCT levels were observed a median of one to two days after enrolment. Patients with severe CAP had highest initial median PCT values (2.4 ng/ml). These patients also showed greater disease severity, organ dysfunction, and mortality than HAP and VAP. This is in concordance with data from Valencia et al., who reported a mortality rate of 37% in CAP patients requiring ICU therapy [27] . Median admission PCTs of 3.4 ng/ml have been observed in patients presenting with CAP in the emergency department [17] . PCT levels were higher, and remained persistently elevated, in non-survivors. Both, initial and maximum PCT values correlated with the maximum SOFA score and were a reasonable predictor of the risk of death within 28 days in these patients. In patients with severe pneumonia, initial PCT measurement allows a risk stratification similar to the APACHE II-score. The data agree with previous observations. In two studies in the emergency department with more than 1,600 patients each, PCT-values < 0.1 ng/ ml in CAP were associated with a low risk of death independent of the clinical risk assessment [17, 18] . PCT was also capable of identifying an unfavorable outcome in CAP patients staying at the ICU [28] . Impact of PCT-assessment is less well investigated in VAP and HAP compared to CAP. Patients with HAP not treated in an ICU have low median PCT values of 0.22 ng/ml [29] . In a single center study conducted in 44 patients with VAP, Duflo et al. found PCT to be significantly elevated in non-survivors: The best cut-off for serum PCT in the non-survivors in the VAP group was 2.6 ng/ml with a sensitivity of 74% and a specificity of 75% [7] . Likewise, Luyt et al. found high median PCT levels of about 3 ng/ml at Day 1 in patients with unfavorable outcomes during the clinical course of microbiologically proven VAP (n = 63) [19] . Interestingly, multivariate analyses further supported that serum PCT levels on days 1, 3, and 7 were strong predictors of unfavorable outcome [19] . We found a significant association between PCT levels and organ dysfunction as assessed by the SOFA score. Similar observations were reported by Meisner et al. [30] and by Schroder et al. in surgical critically ill patients [31] . Hedlund et al. showed that the severity of disease measured by the APACHE II score was strongly associated with admission levels of PCT in 96 adult patients with CAP [32] . In 110 patients with CAP, Boussekey et al. found higher PCT levels in bacteremic patients and/or septic shock patients (4.9 ng/ml vs. 1.5 ng/ml) and in patients who developed infection-related complications (septic shock, multiorgan dysfunction, acute respiratory distress syndrome and disseminated intravascular coagulation) during their ICU stay [33] . The association of PCT with morbidity and mortality may be of clinical importance not primarily for outcome prediction but to monitor success of therapy. Current data support the hypothesis that a drop in PCT levels represents an adequate antimicrobial therapy and may actually define a time point where antibiotic treatment can be safely withdrawn [20, 21] . Recently, this has been Continuous data are given as median (interquartile range) or mean ± standard deviation. CAP, community acquired pneumonia; HAP, hospital acquired pneumonia; ICU, intensive care unit; PCT, procalcitonin; n.s., not significant; SOFA, sequential organ failure assessment; VAP, ventilator associated pneumonia. demonstrated in ICU patients with suspected bacterial infection at admission or during their ICU stay [22] . More than 70% of these patients had pulmonary infections. Unsuccessful source control and poor outcome is associated with persistently elevated PCTs which should negatively affect outcome [14, 34] . Thus, increasing PCT or persistently elevated PCT values should trigger a change in antimicrobial therapy. In this study of severe pneumonia in mechanically ventilated patients, there was no difference in PCT levels between culture positive and culture negative pneumonia. In another study on patients with severe pneumonia as defined by a high Pneumonia Severity Index (PSI), PCT correlated with outcome but could not differentiate between bacterial and nonbacterial etiology of pneumonia [35] . In 72 children with CAP, Moulin et al. found PCT levels > 2 ng/ml in all 10 patients with blood culture positive for S. pneumoniae; PCT concentration was greater than 1 ng/ml in 86% of patients with bacterial infection, with the highest percentage being in those with positive blood culture [36] . This PCT-threshold was more sensitive and specific than CRP, IL-6, or white blood cell count for differentiating bacterial and viral causes of pneumonia. Likewise, Boussekey et al. found higher PCT levels in microbiologically documented CAP (median 4.9 ng/ml vs 1.5 ng/ml if no bacteria were found), but PCT levels could not discriminate between specific bacterial agents [33] . Duflo et al. identified VAP based on a positive quantitative culture of 10 3 colony-forming units/ml or more obtained via a mini-bronchoalveolar lavage. Median PCT values of VAP survivors at baseline were 0.6 ng/ml in this study. This low PCT value questions the validity of currently used VAP diagnostic criteria. Luyt et al. found a similar low PCT of about 0.5 ng/ml in VAP survivors and doubted the usefulness of this parameter for diagnosis of VAP [19, 37] . The 28-day mortality of 8.2% in patients with VAP in our study was very low. The Canadian Critical Care Trials group recorded an overall 28 days mortality rate of 18.7% in a large cohort of patients where VAP was diagnosed using similar criteria as in our study [5] . However, mortality rates between 9.8 and 93.3% have been observed depending on the presence of risk factors such as coexisting diseases, presence of bacteremia, arterial hypotension, or ARDS [38] . The low mortality rate of VAP patients and low PCT-values in the VAP survivors in this study may reflect the uncertainty in correctly diagnosing VAP despite the requirement for a positive Gram stain of respiratory secretion. Although VAP is the most frequent cause of death in hospital for patients with respiratory failure [39, 40] , the diagnosis of VAP is difficult. The optimal invasive procedure for diagnosing HAP or VAP remains poorly defined [9, 10] . Indeed, one study demonstrated that 29% of clinically suspected VAP cases were disproved by autopsy results [41] . In this study, microbiological proof of infection was possible in about 67% of the patients. This is in good agreement with findings in large sepsis trials where microbiological proof was possible in 41 to 51% of the patients with airway infections [42, 43] . It should be noted that the immunoluminometric assay for PCT measurement applied in this study has been replaced today by more modern techniques with a higher accuracy especially in the low range of PCT levels. Such accuracy is a prerequisite when using PCT for antibiotic stewardship [20] . This study was focused on high PCT concentrations for their association with mortality and organ dysfunction. It is unlikely that such a relationship is affected by the type of assay. Measurement of PCT levels in addition to the clinical judgement may offer a solution for this diagnostic dilemma since our data suggest that baseline PCT levels greater than 1.1 ng/ml identify a group of ICU patients with a high risk to develop multiorgan dysfunction followed by death. The quality of mortality prediction was similar to the APACHE II score. These data confirm the observation by Luyt et al., who found a PCT threshold of 1 ng/ml to predict unfavorable outcome [19] . Furthermore, non-survivors showed no decrease in PCT suggesting that pneumonia remained uncontrolled. Assessing adequacy of antimicrobial therapy was not part of the study hypothesis and would have been beyond the scope of this trial. However, PCT measurement offers the possibility of being a marker for monitoring therapeutic success or failure, since successful therapy is associated with a decrease in PCT levels. A PCT guided algorithm has been shown to reduce duration of antibiotic therapy without affecting patients' safety [22, 44] . In patients with severe pneumonia (CAP, VAP, HAP), PCT is associated with the severity of illness and is a good prognostic marker of morbidity and mortality in patients with pneumonia in demand of mechanical ventilation. The severity of illness as reflected by the degree of organ dysfunction may be a more important determinant of PCT levels than the type or cause of pneumonia. • Procalcitonin (PCT) concentrations are associated with the severity of illness in patients with severe pneumonia in demand of mechanical ventilation. • PCT is a good prognostic marker of morbidity and mortality in these patients. • The severity of illness as reflected by the degree of organ dysfunction may be a more important determinant of PCT levels than the type or cause of pneumonia. Abbreviations APACHE II: Acute Physiology and Chronic Health Evaluation II; AUC: area under the curve; BAL: bronchoalveolar lavage; CAP: community acquired pneumonia; CI: confidence interval; GCP: Good Clinical Practice; GCS: Glasgow Coma Scale; HAP: hospital acquired pneumonia; ICU: intensive care unit; IQR: interquartile range; PCT: procalcitonin; PSI: pneumonia severity index; SD: standard deviation; SOFA: Sequential Organ Failure Assessment; ROC: receiver operating characteristic; VAP: ventilator associated pneumonia; WBC: white blood cell count. RB, PL, DA and KR helped to design the study, were responsible for the conduct of the trial, and helped to draft the manuscript. FMB conceived and designed the study and helped to draft the manuscript. All authors read and approved the final manuscript. Competing interests FB received a speaker fee from BRAHMS. ER receives research support from the National Institute of Allergy and Infectious Disease and the Aggennix Corporation and has served as one-time consultant for Aggennix Corporation, Eisai Pharmaceuticals, Idaho Technologies and Astra Zeneca. RB has received research support, consulting fees, and honoraria from BRAHMS and from bioMerieux. DA has received consultant fees from BRAHMS, performed PCT assays for the PCT-7 trial, and had access to equipment and assays by BRAHMS as part of NIH-funded studies. KR has received consultant fees from BRAHMS. FMB has received consultant and speaker fees and grant/research support from BRAHMS. JM, RD, JV, GG and PL declare that they have no competing interests.

Metagenomic Analysis of Fever, Thrombocytopenia and Leukopenia Syndrome (FTLS) in Henan Province, China: Discovery of a New Bunyavirus

Since 2007, many cases of fever, thrombocytopenia and leukopenia syndrome (FTLS) have emerged in Henan Province, China. Patient reports of tick bites suggested that infection could contribute to FTLS. Many tick-transmitted microbial pathogens were tested for by PCR/RT-PCR and/or indirect immunofluorescence assay (IFA). However, only 8% (24/285) of samples collected from 2007 to 2010 tested positive for human granulocytic anaplasmosis (HGA), suggesting that other pathogens could be involved. Here, we used an unbiased metagenomic approach to screen and survey for microbes possibly associated with FTLS. BLASTx analysis of deduced protein sequences revealed that a novel bunyavirus (36% identity to Tehran virus, accession: HQ412604) was present only in sera from FTLS patients. A phylogenetic analysis further showed that, although closely related to Uukuniemi virus of the Phlebovirus genus, this virus was distinct. The candidate virus was examined for association with FTLS among samples collected from Henan province during 2007–2010. RT-PCR, viral cultures, and a seroepidemiologic survey were undertaken. RT-PCR results showed that 223 of 285 (78.24%) acute-phase serum samples contained viral RNA. Of 95 patients for whom paired acute and convalescent sera were available, 73 had serologic evidence of infection, with 52 seroconversions and 21 exhibiting a 4-fold increase in antibody titer to the virus. The new virus was isolated from patient acute-phase serum samples and named Henan Fever Virus (HNF virus). Whole-genome sequencing confirmed that the virus was a novel bunyavirus with genetic similarity to known bunyaviruses, and was most closely related to the Uukuniemi virus (34%, 24%, and 29% of maximum identity, respectively, for segment L, M, S at maximum query coverage). After the release of the GenBank sequences of SFTSV, we found that they were nearly identical (>99% identity). These results show that the novel bunyavirus (HNF virus) is strongly correlated with FTLS.

In May 2007, a county hospital in Xinyang City, Henan Province treated three patients with fever, abdominal pain, bloating, nausea, vomiting, gastrointestinal bleeding, and elevated aminotransferases. The local hospital diagnosed the disease as acute gastroenteritis. A family member of one patient reported the disease to the Henan Center for Disease Control and Prevention (CDC), which sent a team to investigate. The investigation revealed that the disease had the following characteristics: (1) acute onset with fever; (2) low white blood cell and platelet counts; (3) high levels of alanine and aspartate transaminases; (4) positive urine protein. On the basis of these features, the Henan CDC excluded the possibility of gastrointestinal disorders. In order to identify the disease etiology, the Henan CDC team used the above clinical characteristics as the case definition to search for similar cases in local hospitals in this and neighboring counties, while establishing a disease surveillance system that required all medical institutions to report cases that met the above case definition. Altogether, 79 cases were found in 2007 in Henan, with 10 fatalities (case fatality rate, 12.7%). All patients were farmers and resided in mountainous or hilly villages, and many had reported tick bites 7-9 days before illness, further suggesting an infectious etiology. In recent years, patients with similar clinical symptoms were reported with human granulocytic anaplasmosis (HGA; Anaplasma phagocytophilum) in neighboring Anhui province [1] . In 2005, there was an epidemic of Tsutsugamushi (scrub typhus/ Orientia tsutsugamushi) in this area [2] . Clinical investigations, epidemiological analyses, and laboratory testing prompted consideration of rickettsial diseases as possible causes, including HGA, human monocytic ehrlichiosis (HME; Ehrlichia chaffeensis), and Tsutsugamushi disease. Specific methods such as polymerase chain reaction (PCR) and immunofluorescence assays (IFAs) for these pathogens were then used to determine if these cases were attributable to HGA or HME [3, 4] . However, only 18 of 79 (22.7%) patients were positive for A. phagocytophilum based on serology and DNA testing. Thus, the disease was initially considered at least partly caused by A. phagocytophilum, and cases were provisionally diagnosed as suspected HGA based on clinical and epidemiological data [1] [2] [3] [5] [6] [7] . In the 3 years since 2007, 206 suspected cases have been discovered in Henan, but there was only a very low positive rate of A. phagocytophilum confirmation (6 of 206 patients) and no pathogen was isolated. Similar cases were also reported in the mountainous and hilly areas of nearby Shandong, Jiangsu, Hubei and Anhui provinces, indicating that the disease already existed for some time and was widely distributed [7] . We decided to address the possible causative pathogen underlying this infection. On the basis of epidemiological and clinical characteristics, we considered two types of diseases to be possible: rickettsial and arthropod-borne viral disease. Because of the low rates of A. phagocytophilum (rickettsial disease) detection, the research team intensified its virus search to take into account arthropod-borne viruses, including Flaviviridae (Dengue viruses [DENV] , Japanese encephalitis virus [JEV]), Togaviridae (Chikungunya virus, Eastern equine encephalitis virus [EEEV] , Western equine encephalitis virus), and Bunyaviridae (Crimean-Congo hemorrhagic fever virus, Hantaan virus, and Rift valley fever virus) [8] . Specific PCR assays for these viruses were used [9] [10] [11] [12] [13] [14] [15] [16] . However, none of the patients from 2007 to 2010 was positive for these viruses, suggesting a new infectious agent, possibly a virus, remained to be discovered. Thus, the syndrome was considered an emerging infectious disease and was named fever, thrombocytopenia and leukopenia syndrome (FTLS). To identify the etiology, the research team adopted the following strategy: 1) sequencing of randomly amplified cDNA/ DNA from FTLS patient samples using high-throughput Illumina sequencing to specifically explore viral communities present in patients suffering from FTLS, 2) PCR detection of target DNA directly from clinical specimens, 3) viral culture, 4) immunodetec-tion methods, and 5) electron microscopic study of the morphology of the cultured virus. Culture followed by serological and molecular tests is a standard approach for identifying an unknown virus. However, culture of an unknown virus is time-consuming, even taking several years to confirm a novel infection like HIV [17] . Otherwise, virus culture often fails because of the lack of cell lines capable of supporting propagation of viruses (e.g., hepatitis B and C virus). Methods for cloning nucleic acids of microbial pathogens directly from clinical samples offer opportunities for pathogen discovery, thereby laying the foundation for future studies aimed at assessing whether novel or unexpected viruses play a role in disease etiology. Random PCR and subtractive cloning sequencing have identified previously unknown pathogens as etiological agents of several acute and chronic infectious diseases [18, 19] . Recently, high-throughput sequencing approaches have been used for pathogen detection and discovery in clinical samples [20] [21] [22] . We also developed a method for exploring viruses, both known and novel, using highthroughput Illumina sequencing. In this study, high-throughput Illumina sequencing was applied to specifically explore the viral communities in patients with FTLS, using healthy subjects as controls. Here, we provide evidence for the discovery of a novel bunyavirus associated with FTLS through high-throughput sequencing. Subsequent culture of the virus and PCR detection of the specific virus in patient specimens confirmed these findings. Given the serious nature of FTLS, it was decided to handle all clinical specimens and perform all experiments involving live virus in a biosafety level-3 (BSL-3) facility. We studied 285 Henan province patients with FTLS whose samples were submitted to the Henan Province CDC between May 2007 and July 2010. Acute-phase serum samples from all patients were collected. Paired convalescent sera were available from 95 patients. Diagnostic testing of sera for microbial agents possibly related to FTLS Sera were tested by reverse transcription (RT)-PCR, PCR, and/ or indirect IFA serological assays for a number of microbial agents, including A. phagocytophilum, Ehrlichia chaffeensis, Dengue fever virus, Japanese encephalitis virus, Chikungunya virus, Eastern equine encephalitis virus, Western equine encephalitis virus, Crimean-Congo hemorrhagic fever virus, Rift Valley fever virus, Sandfly fever Naples Sabin virus, and Hantavirus [3] [4] [5] [6] [9] [10] [11] [12] [13] [14] [15] [16] 23] . Antigen slides for diagnosis of HGA (A. phagocytophilum) were purchased from Focus Diagnostics (IF1450G, CA, USA). Antigen slides for diagnosis of other pathogens were prepared by our laboratories. Fluorescein isothiocyanate (FITC)-conjugated goat anti-human Initially in 2007, and again between 2008 and 2010, cases of a life-threatening disease with sudden fever, thrombocytopenia, and leukopenia were reported in Henan Province, China. Patient reports of tick bites suggested that this disease could be infectious or tick-transmitted. Many patients were provisionally diagnosed with human granulocytic anaplasmosis (HGA). However, only 24 of 285 (8%) had objective evidence of HGA, suggesting that other pathogens likely contributed to fever, thrombocytopenia and leukopenia syndrome (FTLS). Illumina sequencing was used for direct detection in clinical samples of pathogens possibly associated with FTLS. A novel bunyavirus was found only in samples from FTLS patients. Further epidemiologic and laboratory investigation confirmed that the novel bunyavirus was associated with FTLS. The results illustrate that metagenomic analysis is a powerful method for the discovery of novel pathogenic agents. Combined with epidemiologic investigation, it could assist in rapid diagnosis of unknown diseases and distinguish them from other diseases with similar symptoms caused by known pathogens. IgG (Fc) was purchased from Sihuan Sci-Technics Company (Beijing, China). Equal quantities (100 mL) of acute-phase sera from 10 FTLS patients who had a history of tick bite were pooled and centrifuged at 1000 x g for 10 minutes. The supernatant was collected for DNA and RNA extraction. The same was done for 10 sera from healthy subjects (control). DNA was extracted from 140 mL of each sample using the QIAamp DNA mini Kit (Qiagen, 51304) according to the manufacturer's instructions. DNA was eluted from the columns with 50 mL water containing 20 mg/mL RNaseA. After incubation at 37uC for 15 minutes to eliminate RNA, DNA was used immediately or stored at 280uC. Total RNA was extracted from 140 mL of each sample using the QIAamp viral RNA mini Kit (Qiagen, 52904) according to the manufacturer's instructions. RNA was eluted from the columns with 50 mL of diethyl pyrocarbonate (DEPC)-treated water containing 1 U DNaseI. Samples were incubated at 37uC for 15 minutes to eliminate human DNA, followed by DNase inactivation at 95uC for 10 minutes. RNA was used immediately or stored at 280uC. Random hexamer PCR was carried out in a 25-mL mixture containing 4 mL of cDNA or DNA, 10 mM Tris-HCl (pH 8.4), 50 mM KCl, 2.5 mM MgCl 2 , 100 mM dNTPs, 1 U Taq DNA Polymerase (Promega, M1661) and 100 ng of random hexamer primers containing a linker (5'-GCCGGAGCTCTGCAGAA-TTCNNNNNN-3'). After denaturing at 95uC for 5 minutes, targets were amplified by 45 cycles of 95uC for 30 seconds, 40uC for 30 seconds, 50uC for 30 seconds, and 72uC for 90 seconds. The amplified products were detected by agarose gel electrophoresis. Pure water was used as a negative control. FTLS patient and control genomic DNA/cDNA libraries were constructed according to the manufacturer's instructions (Illumina). In brief, RT-PCR and PCR products were roughly quantified by UV absorption and equal amounts of each sample were mixed. Nucleic acids within the mixtures were sheared by sonication, and fragments in the 150-180-bp range were collected by cutting bands from an agarose gel after electrophoresis. The sheared DNA and cDNA ends were repaired using Klenow DNA polymerase, after which 5' termini were phosphorylated and 3' termini were polyadenylated. The adaptors were added, PCR enrichment was performed, and 150-180-bp fragments were collected for sequencing by the Illumina method. The sequencing procedure was performed according to the manufacturer's instructions (Illumina). In this process, template library DNA was hybridized to the surface of the flow cells and multiple copies of DNA were made to form clusters using the Illumina cluster station. Workflow steps included template hybridization, isothermal amplification, linearization, and final denaturation and hybridization of sequencing primers. Paired-end sequencing (100 cycles) was performed using a four-color DNA Sequencing-By-Synthesis (SBS) technology following the manufacturer's instructions. Short Oligonucleotide Analysis Package (SOAP) was used to handle the large amounts of short reads generated by parallel sequencing [24] . Briefly, after filtering out highly repetitive sequences and adaptor sequences, the overlapping datasets between FTLS and healthy subjects were analyzed by subtracting fragments that mapped to both host genomic-plus-transcript and bacteria databases. The non-redundant reads were mapped onto a virus database downloaded from NCBI (ftp://ftp.ncbi.nih.gov/ genbank/). The resulting alignments were filtered to identify unique sequences by examining alignment (identity $80%) and Evalue scores (e#10 22 ). Filtered unique alignments were examined in the taxonomy database (NCBI) using a custom software application written in Perl (BioPerl version 5.8.5). Unmapped reads were examined in GenBank nucleic acid and protein databases using BLASTn and BLASTx, respectively [25] [26] [27] . Unique alignments were examined in the taxonomy database (NCBI). Sequences without hits were placed in the ''unassigned'' category. Sequences were phylotyped as human, bacterial, phage, viral, or other based on the identity of the best BLAST hit. Considering misannotation and low-complexity for Illumina short reads, sequences assigned to the same virus family were further assembled into contigs with Velvet 1.1.04 (K-mer length = 21; coverage cutoff: default 0; Insert length: PE only; minor contig length: 42) for direct comparison with GenBank nucleic acid databases using BLASTn [28] . Contigs were also blasted with GenBank protein databases using BLASTx [28] . An E-value cutoff of 1610 25 was applied to both BLASTn and BLASTx analyses. Sequences phylotyped as viral were placed in the ''viral'' category. Our mass sequencing data revealed that some sequences showed possible infection with human bocavirus, which belongs to the Parvoviridae family. PCR performed to detect human bocavirus tentatively identified in sera from FTLS patient samples amplified a 291-bp fragment of the NS1 gene, as described previously [29] . Amplified products were detected by agarose gel electrophoresis and sequenced using an ABI 3730 DNA Sequencer. Our mass sequencing data revealed a 168-bp sequence (C361, Accession: HQ412604) indicating the possible presence of a novel virus with closest identity (i.e., lowest E-value) to Tehran virus which belongs to the Phlebovirus genus of the Bunyaviridae family. We developed a PCR strategy based on the 168-bp sequence identified in sera from FTLS patient samples to detect the novel bunyavirus using forward (PF: 5'-GAC ACG CTC CTC AAG GCT CT-3') and reverse (PR: 5'-GCC CAG TAG CCC TGA GTT TC-3') primers designed with Primer3 (Supplementary Figure S1) . PCR was carried out in a 25-mL mixture containing 4 mL of cDNA, 10 mM Tris-HCl (pH 8.4), 50 mM KCl, 2.5 mM MgCl 2 , 100 mM dNTPs, 1 U Taq Pol (Promega, M1661), 0.25 mM forward primer, and 0.25 mM reverse primer. Thermocycling conditions were as follows: 95uC for 4 minutes (denaturation), followed by 35 cycles of 94uC for 30 seconds, 54uC for 30 seconds, and 72uC for 30 seconds. Amplified products were detected by agarose gel electrophoresis and sequenced using an ABI 3730 DNA Sequencer. The Vero E6 cell line (African green monkey kidney cell) was selected for isolation of the novel bunyavirus associated with FTLS because it supports the growth of many bunyaviruses [30, 31] . Vero E6 cell lines were inoculated with six serum samples that contained novel bunyavirus RNA. Each sample underwent at least three cell culture passages in Vero E6 cell line before being considered negative. Medium was replenished on day 7, and cultures were terminated 14 days after inoculation. All cultures were observed daily for cytopathic effect (CPE). Virus-infected cells and uninfected cells were also examined for the novel bunyavirus by RT-PCR at each passage. Vero cell cultures with obvious CPE and containing novel bunyavirus RNA were further analyzed by morphology, genome sequencing, and serology. Cells showing CPE and containing novel bunyavirus RNA were collected for thin-section electron microscopy. After discarding the culture supernatant, virus-infected cells (50 mL) were mixed 1:1 with 4% glutaraldehyde (paraformaldehyde), placed onto Formvar-carbon-coated grids, and stained with 1% methylamine tungstate. Specimens for thin-section electron microscopy were prepared by dehydrating washed cell pellets with serial dilutions of acetone and embedding in epoxy resin. Ultrathin sections were cut on an Ultracut LKBV ultramicrotome, stained with uranyl acetate and lead citrate, and examined under a transmission electron microscope (JEM-1400). The medium from 20 mL of novel bunyavirus-infected Vero E6 cells was centrifuged at 1,000 x g for 10 minutes and then at 4,000 x g for 10 minutes, after which the supernatant was collected. PEG8000 was added to the supernatant at a final concentration of 10% (w/v) followed by centrifugation at 20,000 x g for 2 hours. The pellet was resuspended in 2 mL 16phosphate-buffered saline (PBS) for RNA extraction. Random RT-PCR was performed, and the products (500-1500 bps) were collected and ligated into the pGEM-T vector (Promega, A3600) by incubating overnight at 16uC. Escherichia coli JM109 were transformed with the ligation mixture and cultured on LB agar containing X-gal. White clones were sequenced using an ABI 3730 DNA Sequencer. Contaminating human and extraneous sequences were eliminated using Cross-Match, and the complete sequence was assembled using Phred-Phrap-Consed [32] . Bridge RT-PCR was employed for gap-closure. Phylogenetic analyses were performed using the neighbor-joining method in the MEGA software package, version 4.0.2 [33] . Available nucleotide or protein sequences from known viruses were obtained from GenBank for inclusion in the phylogenetic trees. Selected sequences from GenBank included those with the greatest similarity to the sequence read in question based on BLAST alignments as well as representative sequences from all major taxa within the relevant Bunyaviridae family. To further establish the relationships between the new virus and the members of the Phleboviruses genus, we included all sequences for phleboviruses available in GenBank. Branching orders of the phylograms were verified statistically by resampling the data 1,000 times in a bootstrap analysis with the branch-and-bound algorithm, as applied in MEGA. After successful isolation of the novel bunyavirus, we developed an indirect IFA to detect specific antibodies in patient serum specimens, as previously described [4] . In brief, monolayers of virus-infected Vero E6 cells showing CPE and containing novel bunyavirus RNA were harvested, and one volume of infected cells was mixed with 0.5 volumes of non-infected cells. The mixture was centrifuged at 1,000 x g for 10 minutes, after which cells were resuspended in 16PBS, spotted onto 12-well glass slides, and fixed with acetone for 10 minutes. Sera from patients with FTLS (including 285 acute-phase samples and 95 paired sera), patients with respiratory diseases (80 serum samples), and healthy subjects (50 serum samples) were applied to the cells. Samples (diluted 1:20 in PBS) were screened by first spotting 50 mL of each serum sample per well and incubating for 30 minutes at 37uC. After washing for 10 minutes in PBS, 20 mL of FITC-conjugated goat anti-human IgG (Sihuan Sci-Technics Company, Beijing, China) diluted 1:40 in buffer containing Evans blue was added to each well and incubated for 30 minutes. After washing, slides were mounted in glycerin and examined by immunofluorescence microscopy. A titer of 1:20 was considered positive. The median age of patients was 57.2 years (range, 23-88) and the male-to-female ratio was 1 to 2.27; 219 patients (92.02%) were farmers and 19 (7.98%) were workers or students. Among patients, 52 (21.85%) reported a tick bite within 2 weeks (5-14 days) before the onset of clinical manifestations; the remaining patients did not recall receiving a tick bite. The main clinical features in confirmed patients included sudden onset of fever (.37.5uC 240uC) lasting up to 10 days, fatigue, anorexia, headache, myalgia, arthralgia, dizziness, enlarged lymph nodes, muscle aches, vomiting and diarrhea, upper abdominal pain, and relative bradycardia (Table 1) . A small number of cases suffered more severe complications, including hypotension, mental status alterations, ecchymosis, gastrointestinal hemorrhage, pulmonary hemorrhage, respiratory failure, disseminated intravascular coagulation, multiple organ failure, and/or death. Most patients had a good outcome, but elderly patients and those with underlying diseases, neurological manifestations, coagulopathy, or hyponatremia tended to have a poorer outcome. Laboratory tests showed that confirmed patients characteristically developed thrombocytopenia, leukopenia, proteinuria, and elevated serum aspartate aminotransferase (AST) and alanine aminotransferase (ALT) levels ( Table 2 ). Biochemical tests revealed generally higher levels of lactate dehydrogenase, creatine kinase, AST and ALT enzymes, especially AST. One lane for each of the two sample pools (FTLS and healthy controls), each consisting of 10 samples, was sequenced. The proportions of high-quality sequences in FTLS and control pools were 98.08% (10, 198 ,407/10,397,161) and 97.88% (10,250,809/ 10,472,315), respectively. Unique, high-quality sequence reads were then classified into broad taxonomic groups based on the taxonomy of the most frequent top-scoring BLAST matches for each sequence. Virus sequences constituted 0.0065% (67,969) and 0.0048% (50,677) of the reads in FTLS and control libraries, respectively (Table 3) . After contig assembly, the number of sequences phylotyped as ''viral'' decreased to 3,163 and 2,412, respectively, in the two pools (Table 4) . To screen for possible viruses, we focused exclusively on viruses that were present in patient sera. For detection of known viruses, non-redundant reads were directly aligned with the GenBank database of nucleic acids using BLASTn software. Some sequences from Adenoviridae (human adenovirus), Herpesviridae (herpesvirus), Papillomaviridae (papillomavirus) and Retroviridae (human endogenous retrovirus) were detected in both FTLS patients and healthy subject samples. Some hepatitis B virus (HBV) and human bocavirus sequences were detected only in patient sera, indicating the presence of HBV and human bocavirus infections among these patients. To detect novel viruses, we examined sequence data against the GenBank protein database using BLASTx. An analysis of the deduced protein sequences revealed four different virus families in sera from FTLS patients (Table 4 ). Among these were viruses from Hepadnaviridae, which are not known to cause FTLS; Torque teno virus (TTV) from Anelloviridae, which has been reported to be associated with certain inflammatory states [34] , but is not known to be transmitted by arthropods; and viruses from the Parvoviridae family, including human bocavirus, which could cause febrile illness and signs of FTLS. Although human bocaviruses are not known to be transmitted by arthropods, feline panleukopenia virus, a parvovirus, is strongly suspected to be transmitted by arthropods [35] . All family Parvoviridae sequences detected in FTLS samples were also assembled and their protein sequences deduced. Included among these samples were four fragments, all of which were found to be highly homologous to human bocavirus; one fragment showed the greatest similarity to human bocavirus 2 isolate 53044 (identity = 86%) with the lowest E-value (8610 217 ). Human bocavirus was further detected by PCR in the pool of 10 serum samples, but only one individual sample within the pool tested positive for bocavirus. The final virus family detected in sera from FTLS patients was the Bunyaviridae family, which contains viruses known to cause FTLS after tick bites [8, 30] . All family Bunyaviridae sequences detected in FTLS samples, including 11 fragments (1 S-segment, 2 M-segments, and 8 L-segment fragments), were assembled and their protein sequences were deduced. Among these 11 novel virus fragments was a 168-bp fragment (C361, Accession: HQ412604) of the polymerase gene that showed the greatest similarity to Tehran virus (identity = 36%) with the lowest E-value (3610 27 ). This suggested the presence of a novel virus or a known virus whose genome had not yet been sequenced. To more accurately assess the genetic relationships to known viruses, we constructed a phylogenetic tree using a neighbor-joining method [33] . The result showed that the potentially novel virus clustered with Toscana virus, Uukuniemi virus, and Rift Valley fever virus of the Phlebovirus genus ( Figure 1A ). The pools of 10 serum samples from patients and 10 serum samples from healthy subjects were also screened by PCR for the presence of the novel bunyavirus. All 10 samples from patients tested positive for the novel bunyavirus; however, all 10 samples from healthy subjects tested negative. Thus, we further focused on the virus from the family Bunyaviridae. Using specifically designed RT-PCR primers, we detected viral RNA in 223 of the 285 acute serum samples tested ( Table 5 ). The specificity of the RT-PCR was confirmed by sequencing selected PCR products. None of the 80 sera from patients with respiratory diseases or the 50 sera from healthy subjects was positive using the novel virus-specific RT-PCR. Six acute serum samples that tested positive for the novel bunyavirus by specific RT-PCR were inoculated onto Vero E6 cells, and four virus strains were isolated. The initial CPE analysis showed rounded refractile cells 2-4 days after inoculation. CPE did not progress in the initial cultures, but appeared slightly at 24 hours in subsequent passages (Figure 2A) . RT-PCR revealed the presence of RNA for the novel bunyavirus in all four virus strains, and all isolates reacted with the serum of a convalescent patient in IFA ( Figure 2C ). In addition, electron microscopy showed the presence of virus particles approximately 80-90 nm in diameter-a size compatible with a bunyavirus (Figure 3 ). Virus particles were presumably localized to the Golgi apparatus ( Figure 3 ). Genome sequencing of one isolate (HN01) revealed three segments of negative polarity, single-stranded RNA, including a large segment (L; GenBank HQ642766), a medium-sized segment (M; GenBank HQ642767), and a small segment (S; GenBank HQ642768). The deduced amino acid sequence of the L segment had the highest homology (34%, E value = 3610 2163 ) to RNA polymerase of the Uukuniemi virus of the Phlebovirus genus, whereas the M segment had the highest homology (26%, E value = 8610 255 ) to glycoprotein genes of the Punta Toro virus in the Phlebovirus genus. Of the two proteins encoded by the ambisense S segment, one had the highest homology to nucleocapsid protein (39%, E value = 3610 240 ) of the Rift Valley Fever virus, and the other had the highest homology to nonstructural protein genes (24%, E value = 0.049) of the Punique Virus, both of which belong to the Phlebovirus genus. Collectively, these findings confirm that this virus belongs to the Phlebovirus genus of Bunyaviridae. During the revision of this manuscript, some new sequences of SFTSV (severe fever with thrombocytopenia syndrome) were released. A comparison of these new SFTSV sequences with the sequence of this novel virus showed that they were highly homologous (.99% identity). Using sequences of Phlebovirus available in GenBank, a phylogenetic analysis showed that, although most closely related to the Uukuniemi virus of the Phlebovirus genus (34%, 24%, and 29% of maximum identity, respectively, for segment L, M, S at maximum query coverage), the three genomic segments of the novel virus, along with the SFTSV sequences, were highly divergent ( Figure 1B-1E ). IgG antibodies to the novel bunyavirus were detected in 80 of 285 acute-phase serum samples from patients with FTLS (Table 5) . Of 95 patients from whom paired acute-and convalescent-phase sera were available, 52 had seroconversions and 21 had greater than 4-fold increases in antibody titer to the virus. Six had less than a 4-fold increase in antibody titer to the virus, but all paired sera tested positive. Sixteen patients tested negative to the virus, suggesting that some non-FTLS patients with similar symptoms were included in this study, a situation that is not surprising given that FTLS is a newly emerging disease. The acute-phase sera of four patients from whom the virus was isolated tested negative for IgG antibody to the virus. All convalescent sera obtained 2 months later from the same four patients contained IgG antibody to the virus. None of the 130 sera from patients with respiratory diseases or healthy subjects had detectable antibody. Since 2007, there has been an increase in reported cases of FTLS in Xinyang City, Henan Province. These patients were tentatively diagnosed as having A. phagocytophilum infection. However, only a few (8.4%, 24/285) such patients had evidence for A. phagocytophilum infection, and none of the 285 patients tested positive for the many other pathogens capable of causing similar clinical and laboratory manifestations that were also investigated. These findings suggested novel infectious agents, including viruses. Traditionally, virus culture is very important for identifying an unknown viral infection. Before performing the Illumina sequencing strategy, we attempted viral and rickettsial culture with DH82 and BHK cell lines, but the lack of an obvious CPE led us to initially abandon this approach. Here, mass sequence data obtained by Illumina sequencing revealed four virus families that appeared only in FTLS patient sera. Among these four virus families, viruses from the Parvoviridae and Bunyaviridae families reportedly can cause signs of FTLS and be transmitted by arthropods. However, only one sample from a pool of ten samples tested positive for bocavirus by PCR, suggesting that bocavirus from the Parvoviridae is not likely involved in FTLS. For viruses in the Bunyaviridae family, the incidence of infection is closely linked to vector activity. For example, tick-borne viruses are more common in the late spring and late summer when tick activity peaks. Human infections with certain Bunyaviridae, such as Crimean-Congo hemorrhagic fever virus, are associated with high levels of morbidity and mortality [30] . Considering the tick-bite history of many FTLS patients, we focused on Bunyaviridae family viruses. The entire Bunyaviridae family contains more than 300 members arranged in four genera of arthropod-borne viruses (Orthobunyavirus, Nairovirus, Phlebovirus and Tospovirus) and one genus (Hantavirus) of rodent-borne viruses [30, 36] . The Phlebovirus genus currently comprises 68 antigenically distinct serotypes, only a few of which have been studied. The 68 known serotypes are divided into two groups: the Phlebotomus fever group (the sandfly group, transmitted by Phlebotominae sandflies) comprises 55 members, and the Uukuniemi group (transmitted by ticks) comprises the remaining 13 members. Of these 68 serotypes, eight are linked to disease in humans, including the Alenquer, Candiru, Chagres, Naples, Punta Toro, Rift Valley fever, Sicilian, and Toscana viruses [30] . Phleboviruses have tripartite genomes consisting of a large (L), medium (M), and small (S) RNA segment. In screening for unknown viruses, species hits alone likely carry little weight. Thus, we used all sequences in the family Bunyaviridae for our analysis. A 168-bp fragment of the polymerase gene with the lowest E-value and high sequence identity was used as the sequence of the unknown virus. This virus sequence was detected in all 10 pooled samples, indicating that the virus is involved in FTLS. After detecting a possible novel bunyavirus through highthroughput Illumina sequencing, we inoculated Vero cell lines, which are known to be sensitive to phleboviruses, with sera from six positive patients and were subsequently able to detect the virus by RT-PCR [30, 31] . Although the CPE was modest, RT-PCR confirmed the infection. Genome sequencing was performed and a phylogenetic analysis of the genome sequence showed that this virus clustered into the Phlebovirus branch, but was divergent from other known phleboviruses. These results confirm the novelty of this virus within the Phlebovirus genus of the family Bunyaviridae [36] . Furthermore, virus size and propagation in cells were similar to that of the bunyaviruses. PCR and serological tests were performed to further test the causal link between the new virus and FTLS. Although we have not completely fulfilled Koch's postulates, evidence implicating this new bunyavirus in the outbreak of the disease among patients with FTLS is compelling. In view of the fact that the disease is caused by a novel bunyavirus, and taking into account that the disease was first discovered in Henan (HN), we propose the name "Henan Fever" for the FTLS disease cause by the novel virus (proposed name ''Henan Fever Virus'' [HNF virus]). Since the submission of this manuscript, a bunyavirus was identified as the cause of FTLS in Chinese patients from other regions of China, and the authors have named this virus ''SFTSV'' to indicate that it is the cause of severe fever with thrombocytopenia syndrome [37] . After release of the GenBank sequences referred to in the Yu paper, we compared the sequences of SFTSV with those of FTLSV and found that they were nearly identical (.99% identity). As we first identified the syndrome in 2007 and described the presence of the virus in patients between 2007 and 2010, we suggest that the name ''HNF virus'' should take precedence. The most distinctive feature of the current work includes the use of an unbiased metagenomic approach for viral pathogen discovery that facilitated the rapid creation and implementation of standard culture, serological, and molecular diagnostic approaches. However, there are other differences between the results described here and those reported by Yu et al; notably, we observed slight, but distinctive, CPE in Vero cells. The reason for the failure to observe CPE in Vero cells infected with the ''SFTSV'' bunyavirus [37] , whose genome is nearly identical to that of bunyavirus isolated from our FTLS patients, is unclear. Perhaps this reflects the fact that the ensuing CPE is not dramatic. Alternately, this could indicate the existence of distinct viral strains that vary in pathogenicity, virulence, and possibly even disease manifestations. This is an area of active study in our laboratories. The discovery of this new virus will assist in the rapid diagnosis of this disease and help to distinguish it from other diseases caused by pathogens such as A. phagocytophilum, E. chaffeensis, Crimean-Congo hemorrhagic fever virus, Hantavirus, dengue virus, Japanese encephalitis virus, and Chikungunya virus. Furthermore, the availability of the new virus will facilitate the future development of new therapeutic interventions, such as vaccines and drugs. Figure S1 Sensitivity and dynamic range of real-time PCR in the detection of bunyavirus RNA. To evaluate sensitivity of our RT-PCR, a real-time PCR was performed. Serial dilutions of in vitrobtranscribed bunyavirus RNA sequences were tested. A wide linear range (from 5 copies to 5610 7 copies of control RNA per reaction) was detected in this assay. (TIF)

Clinical aspects and cytokine response in severe H1N1 influenza A virus infection

INTRODUCTION: The immune responses in patients with novel A(H1N1) virus infection (nvA(H1N1)) are incompletely characterized. We investigated the profile of Th1 and Th17 mediators and interferon-inducible protein-10 (IP-10) in groups with severe and mild nvA(H1N1) disease and correlated them with clinical aspects. METHODS: Thirty-two patients hospitalized with confirmed nvA(H1N1) infection were enrolled in the study: 21 patients with nvA(H1N1)-acute respiratory distress syndrome (ARDS) and 11 patients with mild disease. One group of 20 patients with bacterial sepsis-ARDS and another group of 15 healthy volunteers were added to compare their cytokine levels with pandemic influenza groups. In the nvA(H1N1)-ARDS group, the serum cytokine samples were obtained on admission and 3 days later. The clinical aspects were recorded prospectively. RESULTS: In the nvA(H1N1)-ARDS group, obesity and lymphocytopenia were more common and IP-10, interleukin (IL)-12, IL-15, tumor necrosis factor (TNF)α, IL-6, IL-8 and IL-9 were significantly increased versus control. When comparing mild with severe nvA(H1N1) groups, IL-6, IL-8, IL-15 and TNFα were significantly higher in the severe group. In nonsurvivors versus survivors, IL-6 and IL-15 were increased on admission and remained higher 3 days later. A positive correlation of IL-6, IL-8 and IL-15 levels with C-reactive protein and with > 5-day interval between symptom onset and admission, and a negative correlation with the PaO(2):FiO(2 )ratio, were found in nvA(H1N1) groups. In obese patients with influenza disease, a significant increased level of IL-8 was found. When comparing viral ARDS with bacterial ARDS, the level of IL-8, IL-17 and TNFα was significantly higher in bacterial ARDS and IL-12 was increased only in viral ARDS. CONCLUSIONS: In our critically ill patients with novel influenza A(H1N1) virus infection, the hallmarks of the severity of disease were IL-6, IL-15, IL-8 and TNFα. These cytokines, except TNFα, had a positive correlation with the admission delay and C-reactive protein, and a negative correlation with the PaO(2):FiO(2 )ratio. Obese patients with nvA(H1N1) disease have a significant level of IL-8. There are significant differences in the level of cytokines when comparing viral ARDS with bacterial ARDS.

Originating from Mexico and spreading initially in the United States and Canada, a novel influenza A(H1N1) virus infection (nvA(H1N1)) of swine origin spread globally during spring 2009 to mid-February 2010. Rates of hospitalization and death have varied widely according to country [1] . Among hospitalized patients 9 to 31% have been admitted to intensive care units (ICUs) where the rate of death was 14 to 46% [2] [3] [4] [5] [6] . In Romania the pandemic wave lasted from September 2009 to February 2010, reaching a peak in December. The Romanian Ministry of Health reported 7,008 confirmed cases of nvA(H1N1) influenza, the death rate being 1.9%. Primary influenza pneumonia had a high mortality rate during pandemics not only in immune-compromised individuals and patients with underlying co-morbid conditions, but also in young healthy adults [7] . During nvA(H1N1) virus infection, experimental and clinical studies have identified dysregulated systemic inflammation as an important pathogenetic mechanism correlating with severity and progression of the disease [8, 9] . The role of most immune responses in controlling and clearance of H1N1 influenza A or its contribution to severe respiratory compromise is not well known. To and colleagues found higher plasma levels of proinflammatory cytokines and chemokine in the group of patients with acute respiratory distress syndrome (ARDS) caused by viral A(H1N1) influenza, throughout the initial 10 days after symptom onset [8] . Bermejo-Martin and colleagues found that mediators involved in the development of Th17 cells (IL-6, IL-8, IL-9, IL-17), Th1 cells (TNFα, IL-15, IL-12p70) and type II interferon (IFNγ) had high systemic levels in hospitalized patients with nvA(H1N1) influenza [9] . The detrimental or beneficial role of these cytokines in severe illness is not known. The aim of our study was to further investigate the profile of Th1 and Th17 mediators and interferoninductible protein-10 (IP-10), an innate-immunity mediator, as early host response in a group of critical and noncritical hospitalized patients with nvA(H1N1) from Cluj-Napoca, Romania, and to correlate them with the clinical aspects. The study was performed between October 2009 and February 2010 in the ICUs of the Emergency County Clinical Hospital and of the Teaching Hospital of Infectious Diseases, Cluj-Napoca, Romania. Thirty-two patients hospitalized with nvA(H1N1) infection were enrolled in the study: 21 patients with nvA(H1N1)-ARDS, and 11 patients with nvA(H1N1)mild disease. Additionally, 20 patients with bacterial sepsis-ARDS were included and served to compare the cytokine levels between the nvA(H1N1)-ARDS group and the bacterial sepsis-ARDS group. The study protocol was approved by the Ethics Committee for Clinical Research of the University of Medicine and Pharmacy 'Iuliu Hatieganu' Cluj Napoca and the hospital authority. Informed consent was obtained from each patient or their legal representative. The inclusion criteria were age > 16 years, symptoms compatible with influenza and confirmed nvA(H1N1) virus, bacterial severe sepsis with ARDS, and informed consent. The exclusion criteria were age < 16 years, known infection by human immunodeficiency virus, patients with other respiratory viral infections, bacterial sepsis without ARDS-syndrome, and refusal to consent. The control group included 15 healthy volunteers without chronic or acute disease. Data were recorded prospectively by investigators at each hospital. The following data were recorded: age, sex, pregnancy, underlying diseases (chronic obstructive pulmonary disease, asthma, diabetes, chronic heart failure, chronic renal failure, cirrhosis, immunosuppression), obesity defined as body mass index > 30, and the time in days from symptom onset to hospital admission. Hematological, biochemical and microbiological results were included in the database. The extension of lung infiltrates on chest X-ray scan was registered as the number of quadrants involved. The severity and prognosis of the illness was assessed in adults using the Acute Physiology and Chronic Health Evaluation II (APACHE II) score and the Sepsis-related Organ Failure Assessment (SOFA) score. ARDS was defined using the 1994 American-European Consensus Conference definitions [10] . The pulmonary dysfunction score was based on the PaO 2 :FiO 2 ratio, ranging from 0 to 3 where grade 0 represented a ratio less or equal to 250; grade 1, a ratio ranging from 250 to 175; grade 2, a ratio ranging from 100 to 175; and grade 3, a ratio less or equal to 100 [11] . A(H1N1) influenza virus presence was confirmed by testing nasopharyngeal swabs or bronchoalveolar lavage specimens with real-time PCR (commercial kits: Full Velocity SYBR Green QRT-PCR/SuperScript III Platinum One-Step Quantitative RT-PCR Taqman; Invitrogen Corporation, Carlsbad, California, USA) at The National Influenza Centre of Cantacuzino Institute, Bucharest, Romania. In patients with nvA(H1N1)-mild disease, the serum samples were taken on hospital admission. In patients with nvA(H1N1)-ARDS infection, the serum samples were taken on admission to the ICU and 3 days later to determine cytokine kinetics. The installation of ARDS, either viral or bacterial, in the course of the disease determined the time of admission to the ICU. In patients with bacterial sepsis-related ARDS, the serum samples were taken on admission to the ICU. The enrolled patients and the healthy volunteers gave whole blood, which was clotted for 30 minutes at 37°C and stored at -70°C until use. The resulting serum was used for cytokine determination. Seven different serum cytokines (IL-6, IL-8, IL-12p70, IL-15, IL-17, TNFα and IFN-γ) were measured with Luminex 200 (Luminex Corporation, Austin, TX, USA) using a multiplex cytokine kit along with the assay performed in accordance with the manufacturer's instructions (R&D Systems, Minneapolis, MN, USA). Additionally, we used ELISA kits for quantitative determination of the two cytokines IL-9 and IP-10 (Quantikine; R&D Systems). Subjects were stratified into three groups: 11 patients with nvA(H1N1)-mild disease, 21 patients with nvA (H1N1)-ARDS, and 20 patients with bacterial sepsis-ARDS. Descriptive statistics included means and standard deviations or medians and interquartile ranges for continuous variables of normal and non-normal distributions. Clinical and biochemical characteristics and cytokine levels were compared. The Fisher exact test and the chi-square test were used for categorical variables. The Mann-Whitney U test was used for nonparametric variables. The Wilcoxon test (nonparametric test) was used to compare two paired groups. The association between nonparametric variables was determined by the Spearman correlation coefficient (r). Any value of P < 0.05 was considered statistically significant. GraphPad Prism version 5.03 Software for Windows (GraphPad Software, La Jolla, California, USA) was used. A total of 32 patients with confirmed nvA(H1N1) infection and 20 patients with bacterial sepsis-ARDS were enrolled over the study period. Their demographic, co-morbidities and clinical characteristics are presented in Table 1 . Patients in the nvA(H1N1)-ARDS group were significantly older than those in the nvA(H1N1)-mild disease group (median age 42 years vs. 33 years, P = 0.009). Obesity was more common in the nvA(H1N1)-ARDS group. The median interval between onset of illness and admission was 6 days (interquartile range 3.5 to 8.5) in the nvA(H1N1)-ARDS group and 2 days (interquartile range 2 to 3) in the mild disease group (P < 0.001) ( Table 1 ). All the patients with nvA(H1N1) virus infection presented symptoms of acute respiratory viral infection on admission. The median length of hospital stay was higher in the nvA(H1N1)-ARDS group compared with the mild disease group (11 days vs. 6 days, P < 0.001). All patients with nvA(H1N1) virus infection received oseltamivir on admission: the standard dose (150 mg/day) was administered for patients with mild disease, and a higher dose (300 mg/day) was used for nvA(H1N1)-ARDS patients. During the ICU hospitalization, critical patients with influenza virus infection (ARDS) received corticosteroid therapy (hydrocortisone or methylprednisolone). In agreement with our protocol, empirical antibiotics were started on admission. Among 21 patients with nvA(H1N1)-ARDS, four developed acute renal failure requiring renal replacement therapy, two developed secondary bacterial pneumonia and three developed pneumothorax (Table 1 ). Ten patients from the nvA(H1N1)-ARDS group received non-invasive ventilation and 11 patients received mechanical ventilation. Pregnancy was another risk factor for nvA(H1N1)-ARDS infection and ICU admission (3/21 cases; Table 1 ). Two pregnant women were in the third trimester and one was in the second trimester. No underlying disease was noted. The range interval after symptom onset and ICU admission was 3 to 7 days. Caesarean delivery was necessary in two cases. All pregnant women required respiratory support (two invasive and one noninvasive) during hospitalization and all survived. Seven patients died in the nvH1N1-ARDS group. Histopathological changes were similar in all cases: tracheitis, bronchitis with focal squamous metaplasia, necrotizing bronchiolitis, emphysema, extensive diffuse alveolar damage associated with alveolar hemorrhage and marked hyaline membrane formation, fibrosis and granulocyte pulmonary infiltrates. Pulmonary thromboemboli with focal infarcts were observed in three cases. The lymphocyte count was significantly lower in the nvA(H1N1)-ARDS group than in the mild disease group (P = 0.011) ( Table 2 ). Comparing laboratory abnormalities on hospital admission we found that patients with nvA(H1N1)-ARDS were more likely to have elevated levels of serum lactate dehydrogenase, alanine and aspartate aminotransferase (P < 0.001, P = 0.049 and P < 0.001, respectively) than patients with nvA(H1N1)mild disease ( Table 2) . Twenty patients with bacterial sepsis-ARDS were included to compare the cytokine levels in viral and bacterial ARDS. Immune suppression (six patients with cancer) was more common in the bacterial sepsis-ARDS group (P = 0.044). The mean (standard deviation) APACHE II score, SOFA score and PaO 2 :FiO 2 ratio were similar in both groups ( Table 1 ). The leukocyte count, C-reactive protein and procalcitonin levels were higher in the bacterial ARDS group than in the nvA (H1N1)-ARDS group (P = 0.047, P = 0.05 and P < 0.001, respectively) ( Table 2) . The results of the cytokine profile are shown in Figure 1 . At admission, only IL-6, IL-12, IP-10 and TNFα were significantly higher in the mild disease group than in the control group. Except for IL-17 and IFNγ, all cytokine levels were higher in critical patients with nvA (H1N1)-ARDS than in the control group. Compared with the mild disease group, significantly higher levels of IL-6, IL-8, IL-15 and TNFα were found in the nvA (H1N1)-ARDS group (P < 0.001, P < 0.001, P < 0.001 and P < 0.05, respectively). Compared with controls, the levels of IL-6, IL-8, IL-9, IL-15, IL-17, IP-10 and TNFα were significantly elevated in the bacterial sepsis-ARDS group. Levels of IL-8, IL-17 and TNFα were significantly higher in the bacterial-ARDS group versus the nvA (H1N1)-ARDS group (P = 0.05, P = 0.004 and P = 0.011, respectively; Figure 1 ). Patients with pandemic influenza virus (severe ARDS and mild disease) were stratified according to the interval between symptom onset and admission. Levels of IL-6, IL-8, IL-15 and IFNγ were significantly higher in patients with delayed admission, > 5 days after symptom onset (P = 0.006, P = 0.037, P = 0.013 and P = 0.027, respectively) ( Table 3 ). Serum cytokine levels over time (3 days after admission and antiviral treatment) showed a decrease of IL-6, IP-10, TNFα, IFNγ and IL-17 in critical patients with nvA (H1N1)-ARDS (Table 4 ). Serum cytokine levels over time in nvA(H1N1)-ARDS survivors showed a significant decrease of IL-6, IP-10 and TNFα (Table 5 ). In nonsurvivors versus survivors from the nvA(H1N1)-ARDS group, the levels of IL-6 and IL-15 on admission and 3 days after were significantly higher ( Table 6 ). IL-17 was higher in nonsurvivors 3 days after admission (Table 6) . Correlation between cytokine levels and clinical or laboratory characteristics in patients with confirmed nvA(H1N1) infection was determined by Spearman correlation coefficient. We found significant correlation of IL-6, IL-8 and IL-15 levels with C-reactive protein (r = 0.67, P < 0.001; r = 0.5, P = 0.003; and r = 0.48, P = 0.005, respectively), with PaO 2 :FiO 2 ratio (r = -0.556, P = 0.001; r = -0.574, P < 0.001; and r = -0.614, P < 0.001, respectively) and with interval between symptom onset and hospital admission (r = 0.51, P = 0.002; r = 0.41, P = 0.019; and r = 0.48, P = 0.004, respectively). IL-8 was significantly higher (P = 0.013) in obese versus nonobese patients with nvA(H1N1) infection. In this study we presented the cytokine profiles following nvA(H1N1) infection in 32 hospitalized patients (11 mild and 21 severe disease) and the cytokine profiles found in 20 cases of bacterial sepsis. The patients with severe nvA(H1N1) disease were younger than the patients with bacterial sepsis (no statistical significance). Similarly to other study groups, we found that obesity was more common in the nvA (H1N1) ARDS group, suggesting it may be a risk factor for complications and admission to the ICU [2, 5, 6] . Laboratory findings in the same group of patients include lymphocytopenia and elevation in levels of alanine aminotransferase, aspartate aminotransferase, lactate dehydrogenase and creatinine -as in other patient groups with novel influenza virus infection [4, 6] . In contrast, the bacterial-ARDS group presented no lymphocytopenia, lower elevation in serum liver enzymes and higher levels of C-reactive protein and procalcitonin. No significant differences were found between bacterial and viral ARDS groups in SOFA and APACHE II scores at admission. The pulmonary histopathological findings in nvA(H1N1)-ARDS nonsurvivors were similar to other fatal cases of nvA(H1N1) virus infection [12, 13] . Installation of ARDS in the course of the disease was the moment of blood sampling for cytokine measurements. There was a difference regarding the time of symptom onset and hospital admission between the severe and mild groups of nvA(H1N1) disease that could affect the comparison of cytokine levels between the two groups. For this reason we not only compared the cytokine levels between mild and severe disease, but also mixed the patients with nvA(H1N1)-mild and severe disease and compared the level of cytokines according to the interval between symptom onset and admission (first interval 1 to 5 days, second interval 6 to 14 days). We found that not all cytokines had the same behavior against the time of symptom onset and admission. The pattern of immune response in patients with nvA (H1N1) virus infection is incompletely characterized. CD4 + T cells are known to play an important role in the initiation of immune responses by providing help to other cells. T-helper cells could be divided into subsets: Th1, Th2 and Th17. Th1 cells mainly develop following infections by intracellular bacteria and some viruses [14] . The mediators involved in the development of Th1 are IL-12, IFNγ, IL-15, IL-18 and TNFα. IL-12 bridges the early nonspecific innate immunity and the subsequent antigen-specific adaptative immunity [15] . IL-12 was shown to inhibit apoptosis of T cells [16] and of dendritic cells [17] . Alveolar macrophages have a functional IL-12 receptor, and virus-infected macrophages in the presence of IL-12 might be protected from apoptosis limiting viral clearance [18] . Apoptosis of virus-infected cells was shown to be an effective mechanism for viral clearance [19] . Bermejo-Martin and colleagues reported more significant IL-12 results in the critical A(H1N1) group of patients [9] . In our study, IL-12 is significantly higher in the nvA (H1N1)-mild disease group and in the nvA(H1N1)-ARDS group versus the control group and is not significantly higher in the bacterial ARDS group. IL-15 plays a critical role in protecting CD8 + T cells from apoptosis during the contraction phase following microbial infection [20, 21] . The CD8 + T cells surviving in the presence of IL-15 might be pathogenic in lung injury following highly pathogenic influenza A virus infection [22] . IL-15 activates the effector function of memory phenotype CD8 + cells [23] . In our study, IL-15 is significantly higher in the nvA(H1N1)-ARDS group versus the nvA(H1N1)-mild disease group, but without significant difference in the nvA(H1N1)-ARDS versus bacterial-ARDS groups. Similar to our results, IL-15 was a hallmark of critical illness in the Hong Kong and Spanish nvA(H1N1) cytokine studies [8, 9] . IL-15 is significantly higher at admission (P1) and 3 days later (P2) in the nvA(H1N1)-ARDS group for nonsurvivors versus survivors, so it might be pathogenic in lung injury influenza A virus infection. Similarly, To and colleagues found IL-15 significantly higher in critical A(H1N1) patients and very significant in the A(H1N1)-ARDS death group [8] . IFNγ is a cytokine of innate and adaptative immunity. Its major functions are activation of macrophages, differentiation of Th1 from T cells, inhibition of the Th17 pathway and control of intracellular pathogens [24] . Bermejo-Martin and colleagues found high systemic levels of IFNγ in hospitalized patients with nvA(H1N1) [9] . In contrast, in the present study there were no differences between the control and study groups. The IFNγ level over time in the nvA(H1N1) ARDS group was higher at admission than 3 days later, without significant difference between survivors versus nonsurvivors. TNFα is a cytokine of innate immunity. The principal cellular targets and biologic effects include activation of endothelial cells, neutrophil activation, fever, liver synthesis of acute phase proteins, muscle and fat catabolism, and apoptosis of many cell types. In our study, we found highly increased TNFα levels in the nvA(H1N1)mild disease, nvA(H1N1)-ARDS and bacterial ARDS groups compared to the control group. TNFα is significantly higher in nvA(H1N1)-ARDS versus nvA(H1N1)mild disease, with similar results being found by To and colleagues and Bermejo-Martin and colleagues [8, 9] . This cytokine is also significantly increased in bacterial-ARDS versus nvA(H1N1)-ARDS. For the groups of patients with nvA(H1N1), according to the time interval between symptom onset and hospital admission, there were no significant differences found for IL-12 and TNFα levels, but there were significant differences for IL-15 and IFNγ, levels being higher when the time interval was between 6 and 14 days. None of our patients were on oseltamivir medication between symptom onset and admission. Th17 cells are effective in host defense against certain pathogens and tissue inflammation. Th17 mediators for the development of Th17 cells are IL-6, transforming growth factor beta, IL-8, IL-9, IL-17, IL-1 and IL-23. IL-6 is a cytokine of innate immunity, its principal targets being the liver cells, the β cells and the naïve T cells [25] . Despite the apparently beneficial role that macrophages play in controlling early viral replication, several reports have demonstrated a more deleterious effect of these cells in influenza A viral infections by excessive inflammation in the lung attributed to IL-6 and TNFα [26] . In our study, IL-6 is increased in nvA(H1N1)-ARDS versus nvA(H1N1)-mild disease. Similarly, IL-6 and IL-15 constituted a hallmark of critical illness in the Hong Kong and Spanish nvA(H1N1) cytokine studies [8, 9] . In the nvA(H1N1)-ARDS group, the IL-6 serum level is significantly higher at admission than 3 days later. In the same group, IL-6 is significantly higher in nonsurvivors versus survivors at admission and 3 days later, which seems to further contribute to pulmonary damage and death. We found positive correlations between IL-6, IL-15 and IL-8 levels and a longer than 5 days interval between symptom onset and admission, as well as with C-reactive protein, but a negative correlation with the PaO 2 :FiO 2 ratio, indicating the severity of the disease. IL-8 is a chemokine of innate immunity. The chemokine's principal biologic effect is chemotaxis, being a major chemokine for neutrophil activation, and migration into tissues [24] . In our study, IL-8 is highly significant in the nvA(H1N1)-ARDS and ARDS bacterial groups versus the control group, but is not significant in mild disease. In contrast, IL-8 was increased in both critical and noncritical nvA(H1N1) hospitalized patients in the Spanish and Hong Kong studies. In our study, IL-8 is higher in nvA(H1N1)-ARDS versus nvA(H1N1)-mild disease and in bacterial ARDS versus nvA(H1N1)-ARDS. The obese patients with nvA(H1N1) disease had a significant level of IL-8. Plasma IL-8 levels are increased in normoglycemic obese subjects, related to fat mass and the TNFα system [27] . IP-10 is a chemokine of innate immunity, and macrophages and dendritic cells are the principal cell source. We found a higher level of IP-10 in nvA(H1N1)-mild disease, nvA(H1N1)-ARDS and bacterial-ARDS groups versus the control group, and no other differences between groups. In the nvA(H1N1)-ARDS group, the IP-10 level is higher at admission than 3 days after admission because of the survivors' cytokine profile. An increased level of IP-10 was found in the Spanish group as early response to nvA(H1N1) infection in both hospitalized and mild patient disease, as in the present study, while in the Hong Kong group IP-10 was significantly higher in critical patients only. In our study, IP-10 levels in nvA(H1N1)-ARDS nonsurvivors remained higher at admission and 3 days later, being not significantly correlated with the clinical outcome. Emphysema was one of our hystopathological findings and thus it might be speculated that a high level of IP-10 in nonsurvivors could be correlated with emphysema. IP-10 released by lung CD41 and CD81 T cells stimulates alveolar macrophage production of matrix metalloproteinase-12, which digests lung elastin [28, 29] . IL-17 is a cytokine of adaptative immunity. Principal cellular targets include endothelial cells with increased chemokine production and macrophages with increased chemokine and cytokine production. This cytokine's principal biologic effect is proinflammatory [24, 25] . In the present study IL-17 is significantly higher in the bacterial ARDS group versus the control group and is higher in the bacterial ARDS group versus the nvA(H1N1)-ARDS group. No significant differences between nvA(H1N1)-mild disease versus controls and between nvA(H1N1)-ARDS versus controls were found. In the nvA(H1N1)-ARDS group, IL-17 was higher at admission and lower 3 days later. In the Spanish study the IL-17 level was increased in hospitalized noncritical patients, and in the Hong Kong study no differences between groups were found, similar to the present study. IL-9, like IL-6, is a Th2 cytokine that induces differentiation of Th17 cells and has anti-inflammatory properties. IL-9 is a cytokine of current interest associated with allergic Th2 responses and is a key modulator of antiviral immunity [30] . In our study IL-9 is significantly higher in the H1N1-ARDS group versus the control group, and is not significantly increased in mild disease -in contrast to the Spanish study, where IL-9 was increased in both critical and noncritical hospitalized patients. Regarding the behavior of Th17 mediators in nvA (H1N1) groups of patients according to the time interval between symptom onset and admission, there were no differences for IL-9, IL-17 and IP-10 and there were significant differences for IL-6 and IL-8, the levels being higher when the interval was between 6 and 14 days. All our patients with ARDS disease were on corticosteroid treatment, because deficient corticosteroid-mediated downregulation of inflammatory cytokine transcription in ARDS patients is associated with disease progression and mortality. Many studies reported that prolonged corticosteroid treatment was associated with a significant reduction in markers of systemic inflammation [31, 32] . In the present study the blood samples for cytokine measurements were taken at admission for the bacterial-ARDS group of patients, and at admission and 3 days later for the nvA(H1N1) group of patients -for this reason, corticosteroid could not significantly affect cytokine levels. The small number of patients enrolled in the mild disease group is one of our study limitations. Among hospitalized patients with mild flu-like syndrome, only those with risk of severe complications and of secondary outbreaks in the exposed population were sampled for real-time PCR. On the contrary, the laboratory of the National Influenza Centre of Cantacuzino Institute, Bucharest was overwhelmed, being the only centre for influenza PCR diagnosis. Another limitation is the exclusion of children, an important group with nvA (H1N1) virus infection. In our critically ill patients with nvA(H1N1) virus infection we found increased levels of some cytokines: IP-10, TNFα, IL-15, IL-12, IL-6, IL-8 and IL-9. The hallmarks for the severity of the disease were IL-6, IL-15, IL-8 and TNFα. We found a positive correlation of IL-6, IL-15 and IL-8 with the admission delay and C-reactive protein and a negative correlation with the PaO 2 :FiO 2 ratio. The obese patients with nvA(H1N1) disease had a significant level of IL-8. There were significant differences in the level of cytokines when comparing viral ARDS with bacterial ARDS. • In the influenza-related ARDS group, the levels of IL-6, IL-8, IL-9, IL-12, IL-15, IP-10 and TNFα are significantly increased versus the control group. In the bacterial sepsis-ARDS group, levels of IL-6, IL-8, IL-9, IL-15, IL-17, IP-10 and TNFα are also increased versus the control group. When comparing these two groups, the levels of IL-8, IL-17 and TNFα are significantly higher in bacterial ARDS versus viral ARDS, and IL-12 is increased only in viral ARDS whereas IL-17 is increased only in bacterial ARDS. When comparing the mild nvA (H1N1) and critical ARDS influenza A groups, IL-6, IL-8, IL-15 and TNFα are significantly higher in critical ARDS patients being hallmarks of disease severity. • The serum levels of IL-15, IL-6, IL-8 and IFNγ according to the interval between symptom onset and admission in hospitalized nvA(H1N1) patients are significantly higher when this interval is longer than 5 days. • In nonsurvivors versus survivors from the nvA (H1N1)-ARDS group, IL-6 and IL-15 are increased at admission and stay higher 3 days later -which seems to further contribute to pulmonary damage and death. • There is a positive correlation of IL-6, IL-8 and IL-15 levels with C-reactive protein and with > 5-day interval between symptom onset and hospital admission, and a negative correlation with the PaO 2 :FiO 2 ratio. • The obese patients versus nonobese patients with nvA(H1N1) infection have a significant level of IL-8.

Clinical review: Idiopathic pulmonary fibrosis acute exacerbations - unravelling Ariadne's thread

Idiopathic pulmonary fibrosis (IPF) is a dreadful, chronic, and irreversibly progressive fibrosing disease leading to death in all patients affected, and IPF acute exacerbations constitute the most devastating complication during its clinical course. IPF exacerbations are subacute/acute, clinically significant deteriorations of unidentifiable cause that usually transform the slow and more or less steady disease decline to the unexpected appearance of acute lung injury/acute respiratory distress syndrome (ALI/ARDS) ending in death. The histological picture is that of diffuse alveolar damage (DAD), which is the tissue counterpart of ARDS, upon usual interstitial pneumonia, which is the tissue equivalent of IPF. ALI/ARDS and acute interstitial pneumonia share with IPF exacerbations the tissue damage pattern of DAD. 'Treatment' with high-dose corticosteroids with or without an immunosuppressant proved ineffective and represents the coup de grace for these patients. Provision of excellent supportive care and the search for and treatment of the 'underlying cause' remain the only options. IPF exacerbations require rapid decisions about when and whether to initiate mechanical support. Admission to an intensive care unit (ICU) is a particular clinical and ethical challenge because of the extremely poor outcome. Transplantation in the ICU setting often presents insurmountable difficulties.

death in all patients aff ected, and IPF exacerbations constitute the most devastating complication during its course [1] [2] [3] [4] [5] [6] . IPF exacerbations appear more frequently than previously thought and represent a common terminal event [7, 8] . IPF lacks eff ective treatment, and survival is approximately 3 years [2, 6, 9, 10] . Best supportive care constitutes the only attainable therapeutic strategy and includes a more or less eff ective attempt to alleviate symptoms and prevent complications and a far more effi cacious interventional approach consisting of the withdrawal of corticosteroids and immunosuppressants (commonly administered by clinicians) that are ineff ective and harmful [2, 9, 11] . Transplantation is the only thera peutic option [12] . IPF exacerbations represent acute and clinically signifi cant deteriorations of unidentifi able cause, transform ing the slow and more or less steady disease decline [13] to the unexpected appearance of acute lung injury/ acute respiratory distress syndrome (ALI/ARDS) ending in death [6, 14] . Occasionally, IPF exacerbations may present in a previously apparently healthy or minimally symptomatic individual and might represent acute progression of an unsuspected or undiagnosed early IPF [3, 15] . Defi nition criteria include IPF diagnosis, unexplained worsening or development of dyspnea within 30 days, new lung infi ltrates (mainly ground glass upon honeycomb), and exclusion of any identifi able or treatable cause of lung injury [6] . Surgical lung biopsy per se constitutes a risk factor for their development [16] but, when performed for the investigation of the etiology of exacerbations or in autopsies, discloses a histological picture of diff use alveolar damage (DAD), which is the ARDS tissue counterpart, upon usual interstitial pneumonia (UIP), which is the IPF tissue equivalent [4, 8, [17] [18] [19] . In IPF, anachronic and reiterative epithelial injury and loss of the alveolar-capillary integrity constitute the initial event and 'the point of no return' that trigger aberrant repair pathways leading to inappropriate, progressive, and heterogeneous lung scarring (UIP) [20] [21] [22] . DAD upon UIP might represent massive epithelial and endothelial injury of the lung areas yet preserved from Abstract Idiopathic pulmonary fi brosis (IPF) is a dreadful, chronic, and irreversibly progressive fi brosing disease leading to death in all patients aff ected, and IPF acute exacerbations constitute the most devastating complication during its clinical course. IPF exacerbations are subacute/acute, clinically signifi cant deteriorations of unidentifi able cause that usually transform the slow and more or less steady disease decline to the unexpected appearance of acute lung injury/acute respiratory distress syndrome (ALI/ARDS) ending in death. The histological picture is that of diff use alveolar damage (DAD), which is the tissue counterpart of ARDS, upon usual interstitial pneumonia, which is the tissue equivalent of IPF. ALI/ ARDS and acute interstitial pneumonia share with IPF exacerbations the tissue damage pattern of DAD. 'Treatment' with high-dose corticosteroids with or without an immunosuppressant proved ineff ective and represents the coup de grace for these patients. Provision of excellent supportive care and the search for and treatment of the 'underlying cause' remain the only options. IPF exacerbations require rapid decisions about when and whether to initiate mechanical support. Admission to an intensive care unit (ICU) is a particular clinical and ethical challenge because of the extremely poor outcome. Transplantation in the ICU setting often presents insurmountable diffi culties. scarring [9, 23] . Putative initiators of IPF include viruses, cigarette smoke, gastroesophageal refl ux, and occupational exposure to wood and metals [24, 25] . Aging, by reducing effi ciency in repairing damage, represents a cofactor [26] . Th e development of DAD upon UIP may relate to a clinically occult infection [14, 27] , aspiration, or a distinct pathobiological manifestation of IPF [6] . 'Treatment' with high-dose corticosteroids with or without an immunosuppressant proved ineff ective and represents the coup de grace for these patients [8] . IPF exacerbations require rapid decisions about when and whether to initiate mechanical support. However, the con sideration of admission to an intensive care unit (ICU) is a particular clinical and ethical challenge because of poor outcome [28] [29] [30] [31] [32] . Transplantation in this setting presents insurmountable diffi culties. Th e incidence of IPF exacerbations varies greatly between studies (from 8.5% to 60%) mainly because of diff erences in their design [3] [4] [5] [6] 8, 14, 16, 19, 28, 29, [31] [32] [33] [34] [35] : (a) case series and retrospective cohorts [4, [17] [18] [19] , (b) randomized controlled trials of specifi c treatments for IPF [3] , (c) autopsy reviews [8, 16, 33] , and (d) retrospective reviews of ICU admissions [28, 29, 31, 32] . Discrepancies in reported frequen cies should be attributed to the diffi culty in strictly respecting the defi nition criteria especially concerning symptom duration (less than 4 weeks) and the defi nite exclusion of infection [3, 6, 36] . IPF exacerbations do not appear to be linked to disease duration, functional derange ment, age, gender, or smoking history [4, 29] , although further studies are necessary to confi rm early development as well as lack of association with immunosuppression [37] . Exacerbation mortality approaches 100%, questioning the need for ICU admission [2] [3] [4] [5] [6] 8, 14, 16, 19, 28, 29, [31] [32] [33] [34] . Th e defi nition of IPF exacerbations 'after excluding identi fi able causes of lung injury' implies that in 'idiopathic' pulmonary fi brosis, 'idiopathic' exacerbations occur [3, 4, 6] . However, in clinical practice, when such a patient is referred to the emergency department (ED), the attending clinician has to face one of three clinical scenarios [38] (Figure 1 ). Th e fi rst scenario is the case in which the physical evolution of IPF comes to the fi nal end in which spontaneous breathing becomes unsup portable [39] (Figures 1a and 2) . In this scenario, the exclusion of 'identifi able-treatable causes of lung deterior ation' is demanding, but the only option attainable is palliation. Th e second scenario refers to 'true' IPF exacerbation that brings the patient to the ED (Figures 1b and 3 ). In this case, after admission to the hospital ward, the patient usually becomes unable to maintain spontaneous breathing within hours or very few days, often not enough time for the extensive work-up required to identify treatable factors of deterioration, and needs ventilatory support and ICU transfer [5, 7, 8] . Th e third scenario refers to the admission to the hospital ward of an IPF-deteriorated patient because of reversible causes either aff ecting the lung or not; in this case, early identifi cation of the precipitating factor(s) and their prompt treatment are imperative (Figures 1c and 4) . Nevertheless, borders between the above scenarios are unclear in routine clinical practice since exacerbations occur as a spectrum rather than a clearly defi nable event. However, even after the exclusion of any identifi able and treatable factor(s) inducing IPF exacerbations, the most important etiologic hypothesis remains that of a clinically occult infection that precipitates an already UIP-scarred lung into DAD [6] . For several reasons, viruses are the best etiologic candidates: (a) Epstein-Barr, cytomegalovirus, hepatitis C, herpes simplex, parvovirus B19, torque teno, and especially herpes viruses 7 and 8 have been implicated in IPF pathogenesis [40] [41] [42] ; (b) fl ulike illness heralds the onset of exacerbations, and IPF mortality seems to peak in winter time and coincides with the peak of viral respira tory infections [43] ; (c) in the mice pulmonary fi brosis experimental model, gamma herpesvirus induces exacer ba tions [44] as well as other viruses in vivo [45] ; and (d) latent lung viral infections may reactivate under immuno suppression commonly used by clinicians [41] . Th erefore, in IPF, viruses may act as both initiators and exacerbators because of their formidable ability to induce ARDS [46] . Besides viruses, microbials in traction bronchiectases/ bronchiolectases are equally strong candidates. Bronchiec tases are among the most common of the whole spectrum of lesions that characterize the architectural distortion in IPF. Interleukin-8, neutrophils, and alphadefensins are increased or activated in stable or exacerbated patients with IPF [47, 48] and possibly play a role in triggering ARDS. In addition, immunosuppressive treatment certainly increases suscep ti bility to microbials. Accordingly, further considerations have to be made. ALI/ARDS, acute interstitial pneumonia (AIP), and IPF exacerbations have common clinical, physiological, imaging, and histopathology features, and it is incon ceivable that they do not also have common etiopathogenetic mechanisms ( Figure 5 ). ALI/ARDS develops by diff erent insults to the lung, and the mainstay of its treatment is provision of excellent supportive care and etiologic manage ment of the underlying cause [46] . AIP is precisely an ARDS of 'unknown cause' , and no specifi c clinical clues to diff erentiate between 'known and unknown cause' ARDS exist [49] . Criteria for the diagnosis of AIP are the same as in IPF exacerbations with the exception of the 'incubation' time (2 months instead of 4 weeks) and the prerequisite of normal chest roent genogram. AIP, incomprehensibly, is included among the idiopathic interstitial pneumonias (IIPs) and probably should be added to the list of unknown cause ALI/ARDS, although some believe that AIP may represent a fulminant presentation of IIP secondary to imprecise autoimmune factors [1, 2] . Although there are no controlled trials of specifi c treatment, intensive immunosuppression has been the mainstay of treatment (usually under the coverage of several broad-spectrum antimicrobials, although this is not always stated) because of the inclusion of AIP among the IIPs [50] . However, AIP mortality approaches that of IPF exacerbations, and the provision of excellent suppor tive care and further search of underlying causative factors and adequate treatment seem more logical. In stable IPF (in contrast to other pneumonias), lung damage is not resolved by restitutio ad integrum. IPF exacerbations characterized by DAD upon UIP may represent the acute response of scarred and irreparably damaged lung. Epithelial cell apoptosis involves and is considered to be among the main pathogenetic mechanisms in the development of any DAD [51, 52] . Th erefore, it seems incoherent that DAD, which is the common denominator of all ALI/ARDS, AIP, and IPF exacerbations and which develops upon diff erent histology substrates (UIP in IPF exacerbations, normal lungs in AIP, and normal or diseased lungs in ARDS), presents at diff erent time intervals (7 days for ARDS [46, 53] , 4 weeks for IPF exacerbations [7] , and 2 months for AIP [1] ) and requires diff erent pharmacologic approaches, which proved certainly fatal in AIP and in IPF 'true' exacerbations. Early, accurate, and secure diagnosis is critical in IPFexacer bated patients with reversible precipitating [7] . For details about laboratory tests and blood/sputum/bronchoalveolar lavage (BAL) tests, see the 'Clinical and laboratory assessment' section. Cardiac echo, cardiac echocardiography; CTPA, computed tomography pulmonary angiography; HRCT, high-resolution computed tomography; ICU, intensive care unit; PE, pulmonary embolism; PH, pulmonary hypertension; PNX, pneumothorax; proBNP, pro-brain natriuretic peptide. factor(s) (Figure 1 ) [2] . Investigation into medical history should focus on smoking habits, toxic exposures, prescribed medications, immunosuppression, and signs and symptoms of potentially undiagnosed autoimmune rheumatic disease [2, 54, 55] . Physical examination frequently reveals tachypnea, cyanosis, digital clubbing, bilateral inspiratory crackles, and lower extremity edema. In the most severe cases, the patient may be obtund or comatose because of severe hypoxemic and potentially hypercapnic respiratory failure. Th e presence of arrhyth mias, chest pain, hemoptysis, or hemodynamic instability should guide the physician to an overlapping or alternative diagnosis such as acute coronary syndrome or pulmonary embolism. Chest roentgenograms, including past imaging data, may help to orientate the clinician toward the identifi cation of the causative agents of the exacerbation. Computed tomo graphy pulmonary angiography is mandatory to exclude pulmonary embolism, and high-resolution computed tomography (HRCT) may document extension High-resolution computed tomography shows mild reticulation. (d) Roentgenogram of the patient 24 months after diagnosis demonstrates worsening of the reticular pattern superimposed on a ground-glass pattern. The patient was admitted with severe breathlessness and productive cough. Her symptoms were severely aggravated in the last 9 months and she was hospitalized many times. She had received corticosteroids and mycophenolate mofetil, which were discontinued months prior to this roentgenogram because of lower respiratory tract infections. At the time of the roentgenogram, she was receiving only proton pump inhibitors. She deteriorated further despite best supportive care and died while on palliation treatment. Our putative diagnosis was IPF progressing to the fi nal end. of honey combing or other lung comorbidities (Figure 1 ). Echo cardio graphy may also be useful. When early undiagnosed IPF presents with fulminant respiratory insufficiency and ARDS [3, 29] , honeycombing with bibasilar and subpleural distribution on HRCT [1] can establish the diagnosis of IPF exacerbation and diff eren tiate defi nitely from AIP [49] . In IPF exacerbations, HRCT reveals new bilateral ground-glass abnormalities or consolidations (or both) upon UIP pattern [6] . A ground-glass pattern, especially if extensive, is not a feature of stable IPF, and its rapid development away from areas of fi brosis heralds DAD. Akira and colleagues [18, 56] have proposed a classifi cation of acute exacer bations of IPF on the basis of three ground-glass and consolidation computed tomography patterns that appear to have prognostic implications: (a) peripheral, (b) multifocal, and (c) diff use, though others did not confi rm a similar assumption [57] . Since no laboratory test is specifi c to IPF exacerbations, most tests are performed to exclude treatable causes of deterioration and to document the severity of the exacerbations. Th e standard laboratory work-up should include all necessary tests for the investigation of a critically ill patient with impeding ALI/ARDS of unknown etiology. ALI and ARDS criteria (arterial partial pressure of oxygen/fraction of inspired oxygen [PaO 2 /FiO 2 ] of less than 300 and less than 200, respectively) should prepare the clinician for the possibility of mechanical support. Accurate diagnosis in IPF exacerbations requires bronchoalveolar lavage (BAL) to exclude infection or alternative diagnoses; BAL is best performed before mechanical support or immediately afterwards [2] . Perform ing lung biopsy could be justifi able when facing a disease with grave prognosis but bears an increased risk for postsurgical complications and should be individual ized to each patient [2] . IPF exacerbations lack an eff ective treatment. Intensive immunosuppression proved harmful and fatal [2] . Patients presenting with IPF exacerbations must be managed in centers specializing in interstitial lung diseases (ILDs) with the availability of various specialties and departments such as a respiratory ward with a respiratory ICU/highdependency unit (RICU/HDU), an ICU, and possibly a cardiothoracic transplantation center on a 24-hour basis. Lung transplantation constitutes a treat ment option for IPF 'true' exacerbations [2, 38] but faces insurmountable diffi culties, even in specialized centers. Management depends on the clinical scenario ( Figure 1 ). In case of progression to the fi nal end (Figure 1a) , palliation is more appropriate [2] . Noninvasive ventilation (NIV), by decreasing breathing work, is considered a major palliative option that, together with best supportive care, may help to reduce patient discomfort and permits management in an RICU [58, 59] . Patients with 'true' IPF exacerbations (Figure 1b) , in which the diagnostic approach fails to identify a possible infective etiology, must continue to receive empirical antimicrobial therapy that takes into consideration factors such as immunosuppression, previous colonization, BAL timing, onset of mechanical support, and results of obtained cultures [2] . 'Specifi c' therapies for 'true' IPF exacerbations until now have consisted of highdose intravenous corticosteroids plus an immunosuppressant [2] . However, Cochrane reviews for the effi cacy of these therapies concluded that there is no evidence for any benefi t of both cortico steroids and immunosuppressants in IPF [60, 61] . Besides, both progression of fi brosis on native lung in single-lung transplant patients and IPF 'true' exacerbations have been described in the heavily immunodepressed trans planted patient [10] . NIV may also help to wean the very few survivors from the IPF exacerbations and also constitutes the bridge to transplantation [62] . To promptly recognize and treat reversible precipitating factors implicated in IPF exacerbations (Figure 1c) , recovery in the RICU/HDU or (in case of multiorgan failure) in the ICU is mandatory [63] . An IPF patient is referred to the ICU for severe acute respiratory failure as a consequence of the clinical scenarios (mentioned above) that may lead to ventilatory support (Figure 1 ). Progression to the fi nal end reaches a point at which spontaneous ventilation in no longer possible (Figure 1a ). ICU admission of these patients, because of the poor outcome, should be avoided [2] ( Figure 2 ). In 'true' IPF exacerbations (Figure 1b) , ventilatory support and ICU transfer buy time and could have some infl uence on fi nal outcome in specifi c patients. Unfortunately, in the vast majority, this does not happen, and the mortality of this patient population is high, higher even than that predicted by the usual clinical score HRCT shows, at the lung bases, ground-glass opacities upon extensive peripheral thickening of intralobular septa. The patient was a 65-year-old male with IPF and initiated treatment with high doses of corticosteroids. (c) Four months later, HRCT denotes diff use ground-glass with irregular reticulation. Note the extensive lipomatosis of the mediastinum due to chronic steroid use. Owing to deterioration of dyspnea, he was admitted to another hospital, where bronchoalveolar lavage (BAL) was performed and the immunosuppressive treatment was intensifi ed. A few weeks later, he was admitted to our department with respiratory failure, severe corticosteroid-related myopathy, diabetes mellitus, severe dyspnea, and purulent sputum. Clinical examination disclosed herpes simplex virus keratitis in the left eye, and BAL cultures grew positive for Pseudomonas aeruginosa. Corticosteroids were tapered, and antimicrobial and antiviral treatment was initiated. Both eye and lower respiratory tract infections subsided, and the patient was discharged home a few weeks later. (d) Eighteen months after the exacerbation, the groundglass opacities completely resolved as did the lipomatosis of the mediastinum. The patient is still alive and at home. [2, 31] (Figure 3 ). Admission of an IPF patient to the ICU because of reversible causes either aff ecting the lung or not (Figures 1c and 4 ) bears better prognosis, but special attention should be paid to avoid further complications. Th e complexity of the above scenarios underscores the importance of good communication between referring and ICU physicians. So far, the studies of IPF patients in the ICU have had many limitations (Table 1) [4, 28, 29, 31, 32, 34, 59, 64] . Th ese studies are usually retrospective and single-centered and include limited numbers of patients. In addition, most of these studies include all IPF patients admitted to the ICU for respiratory failure regardless of etiology, the proportion of patients with confi rmed diagnosis is variable, the ventilator parameters are usually not reported, and the pharmacologic therapy demonstrates a signifi cant diversity. Th e only common parameter is the conclusion: the prognosis of ventilated IPF patients is disappointing [2] . Given these results, what may be the goals of ICU support for a patient with an IPF exacerbation? Although defi nite conclusions cannot be drawn, there is a general feeling that mechanical ventilation and intensive support do not have a signifi cant eff ect on outcome [2] . Could this be due to the disease itself, ventilator-induced lung injury, complications of intensive support (sepsis, critical care myoneuropathy, or ventilator-associated pneumonia), or a combination of the above? Only assumptions can be made, and patients (at an earlier stage) and relatives as well as physicians outside of the ICU before or at admission should become aware of the poor prognosis. Th is does not mean that IPF patients with acute respiratory failure should be denied admis sion; in many hospitals, the ICU is the right place to perform in a safe and timely fashion the necessary extended investigation to exclude reversible causes of deterioration in these patients. Ventilating a patient with an IPF exacerbation is a diffi cult and demanding task, and no 'cookbook' prescriptions can make the work easier for the intensivist. Th e evidence for the best ventilator strategy applying to an IPF exacerbation is extremely scarce, and the eff ect of ventilatory management on outcome has not been system atically assessed; therefore, every suggestion is based on theoretical principles and pathologic data that are characterized mainly by extended DAD [7] . Recently, Bates and colleagues [65] introduced the concept of percolation, according to which the progression of parenchymal lung disease can suddenly reach a threshold that dramatically alters the mechanical properties of the lung. IPF exacerbations that require ventilator support could be an example of crossing this percolation threshold. Under these circumstances, mechanical ventila tion could represent a second hit for the lung parenchyma, further deteriorating the mechanical properties of lung parenchyma and introducing a vicious cycle that ends in death. Mechanical ventilation with conventional volumes (8 mL/kg) in patients without lung injury can induce severe surfactant impairment, and sustained plasma cytokine production has been demonstrated in patients without ALI ventilated with conventional tidal This non-proportional fi gure denotes the incoherence of the clinical signifi cance of acute respiratory distress syndrome (ARDS), acute interstitial pneumonia (AIP), and idiopathic pulmonary fi brosis (IPF) exacerbations in which DAD, despite being the common denominator, develops upon diff erent histology substrates (UIP in IPF exacerbations, normal lungs in AIP, and normal or diseased lungs in ARDS) and, according to current defi nitions, presents at diff erent time intervals: 7 days for ARDS, 4 weeks for IPF exacerbations, and 2 months for AIP. This incoherence led also to a diff erent pharmacologic approach, which proved to be unsuccessful at least in AIP and in IPF true exacerbations. ALI, acute lung injury. volumes (10 mL/kg) compared with those ventilated with low tidal volumes (6 mL/kg) [66, 67] . So it should not be surprising, although it may be very diffi cult to prove, that the employment of traditional tidal volumes in patients with IPF exacerbations would be detrimental given that their lungs are characterized by extended parenchymal alterations, severe inhomogeneity, and decreased compliance even prior to initiation of mechanical ventilation. Especially the inhomogeneity of the lung parenchyma could cause severe overinfl ation of the 'healthy' lung units with higher compliance and jeopardize the 'healthy' parenchyma left. A ventilation strategy employing low tidal volumes (4 to 6 mL/kg ideal body weight), such as that used for patients with ARDS, seems prudent and is advised by many experts [68] . Positive end-expiratory pressure (PEEP) should be used moderately because of the aforementioned risk of overinfl ation of intact lung units. Fernández-Pérez and colleagues [64] demonstrated that high PEEP was independently associated with increased mortality in chronic ILD [64] . In the same context, there is no place for recruitment or prone position [69] . Given that intubated patients with IPF exacerbations require high-minute volume because of increased dead space, the respiratory frequency should be increased to the maximum acceptable rate and the target of a normal PaCO 2 (arterial partial pressure of carbon dioxide) should be abandoned. Th is high respiratory rate might require the use of heavy sedation and quite often paralysis, and care should be given to avoid auto-PEEP [70] . Th e eff ect of prolonged sedation and paralysis on the neuromuscular function of these patients, who have often been administered steroids for a long time, is an unavoidable cost. Th e earliest possible interruption of sedation will facilitate weaning provided that gas exchange and lung mechanics have improved. NIV has some theoretical advantages in IPF patients and has been used extensively in cases of acute respiratory failure to avoid intubation. Unfortunately, the studies about its use have the same methodological problems as those for the invasive ventilation studies mentioned previously, and no fi rm conclusions can be drawn. Th ere are two things that make NIV more 'attractive' in this setting: the almost absolute mortality that invasive mechanical ventilation carries and the avoidance of intubation and ventilation risks (aspiration, ventilator-associated pneumonia, and ventilatorassociated injury). Th e problem is that in most cases the excessive work of breathing associated with IPF exacerbation cannot be managed eff ectively by NIV. Extracorporeal membrane oxygenation could represent a valuable adjunct to conventional treatment for selected cases of IPF. Limited availability, high cost, complicated technology, and increased rates of complications have been the most important factors limiting its use so far [71] [72] [73] . In the setting of IPF therapeutics, it has been used mainly as a bridge to transplantation [74] . Transplantation represents the fi nal line of defense for the IPF patient and is the only therapy with a proven survival benefi t. Early referral (even at the time of diagnosis) to a lung transplant center is mandatory [75] because of the prolonged waiting-list time, which sometimes exceeds the patient's life expectancy. IPF exacerbations constitute the most devastating compli cation of IPF. Diff erent and hard-to-diff erentiate clinical scenarios may reproduce the hallmark of their defi nition: subacute/acute deterioration of dyspnea and bilateral chest infi ltrates, corresponding in 'true' IPF exacerbations, to a tissue pattern of DAD upon UIP. Also, ALI/ARDS and AIP present DAD. Intensive immunosuppression proved ineff ective and represents the coup de grace for these patients. Provision of excel lent supportive care and the search for and treatment of the 'underlying cause' remain the only options. Th e unravelling of Ariadne's thread continues. Abbreviations AIP, acute interstitial pneumonia; ALI, acute lung injury; ARDS, acute respiratory distress syndrome; BAL, bronchoalveolar lavage; DAD, diff use alveolar damage; ED, emergency department; HDU, high-dependency unit; HRCT, high-resolution computed tomography; ICU, intensive care unit; IIP, idiopathic interstitial pneumonia; ILD, interstitial lung disease; IPF, idiopathic pulmonary fi brosis; NIV, non-invasive ventilation; PEEP, positive end-expiratory pressure; RICU, respiratory intensive care unit; UIP, usual interstitial pneumonia. The authors declare that they have no competing interests.

Influence of genetic variability at the surfactant proteins A and D in community-acquired pneumonia: a prospective, observational, genetic study

INTRODUCTION: Genetic variability of the pulmonary surfactant proteins A and D may affect clearance of microorganisms and the extent of the inflammatory response. The genes of these collectins (SFTPA1, SFTPA2 and SFTPD) are located in a cluster at 10q21-24. The objective of this study was to evaluate the existence of linkage disequilibrium (LD) among these genes, and the association of variability at these genes with susceptibility and outcome of community-acquired pneumonia (CAP). We also studied the effect of genetic variability on SP-D serum levels. METHODS: Seven non-synonymous polymorphisms of SFTPA1, SFTPA2 and SFTPD were analyzed. For susceptibility, 682 CAP patients and 769 controls were studied in a case-control study. Severity and outcome were evaluated in a prospective study. Haplotypes were inferred and LD was characterized. SP-D serum levels were measured in healthy controls. RESULTS: The SFTPD aa11-C allele was significantly associated with lower SP-D serum levels, in a dose-dependent manner. We observed the existence of LD among the studied genes. Haplotypes SFTPA1 6A(2 )(P = 0.0009, odds ration (OR) = 0.78), SFTPA2 1A(0 )(P = 0.002, OR = 0.79), SFTPA1-SFTPA2 6A(2)-1A(0 )(P = 0.0005, OR = 0.77), and SFTPD-SFTPA1-SFTPA2 C-6A(2)-1A(0 )(P = 0.00001, OR = 0.62) were underrepresented in patients, whereas haplotypes SFTPA2 1A(10 )(P = 0.00007, OR = 6.58) and SFTPA1-SFTPA2 6A(3)-1A (P = 0.0007, OR = 3.92) were overrepresented. Similar results were observed in CAP due to pneumococcus, though no significant differences were now observed after Bonferroni corrections. 1A(10 )and 6A-1A were associated with higher 28-day and 90-day mortality, and with multi-organ dysfunction syndrome (MODS) and acute respiratory distress syndrome (ARDS) respectively. SFTPD aa11-C allele was associated with development of MODS and ARDS. CONCLUSIONS: Our study indicates that missense single nucleotide polymorphisms and haplotypes of SFTPA1, SFTPA2 and SFTPD are associated with susceptibility to CAP, and that several haplotypes also influence severity and outcome of CAP.

Community-acquired pneumonia (CAP) is the most common infectious disease requiring hospitalization in developed countries. Several microorganisms may be causative agents of CAP, and Streptococcus pneumoniae is the most common cause [1] . Inherited genetic variants of components of the human immune system influence the susceptibility to and the severity of infectious diseases. In humans, primary immunodeficiencies (PID) affecting opsonization of bacteria and NF-Bmediated activation have been shown to predispose to invasive infections by respiratory bacteria, particularly S. pneumoniae [2] . Conventional PID are mendelian disorders, but genetic variants at other genes involved in opsonophagocytosis, with a lower penetrance, may also influence susceptibility and severity of these infectious diseases with a complex pattern of inheritance [3] . In the lung, under normal conditions, microorganisms at first encounter components of the innate immune response, particularly alveolar macrophages, dendritic cells and the lung collectins, the surfactant protein (SP)-A1, -A2 and -D. SP-A1, -A2 and -D belong to the collectin subgroup of the C-type lectin superfamily, and contain both collagen-like and carbohydrate-binding recognition domains (CRDs) [4] . Upon binding to pathogen-associated molecular patterns (PAMPs), SP-A and SP-D enhance the opsonophagocytosis of common respiratory pathogens by macrophages [5, 6] . Mice rendered SP-A or SP-D deficient exhibit increased susceptibility to several bacteria and viruses after intratracheal challenge [7] [8] [9] . SP-A1, -A2 and -D also play a pivotal role in the regulation of inflammatory responses [4, 10, 11] and clearance of apoptotic cells [4, 12, 13] . In mice, SP-A and SP-D have been shown to be nonredundant in the immune defense in vivo [9] . The human SP-A locus consists of two similar genes, SFTPA1 and SFTPA2, located on chromosome 10q21-24, within a cluster that includes the SP-D gene (SFTPD) [11] . The nucleotide sequences of human SFTPA1 and SFTPA2 differ little (96.0 to 99.6%) [14] . Single nucleotide polymorphisms (SNP) at the SFTPA1 codons 19, 50, 62, 133 and 219, and at the SFTPA2 codons 9, 91, 140 and 223 have been used to define the SP-A haplotypes, which are conventionally denoted as 6A n for the SFTPA1 gene and 1A n for the SFTPA2 gene (see Table E1 in Additional File 1) [15] . Variability at the SFTPD gene has been also reported. Particularly, the presence of the variant amino acid (aa)-11 (M11T) has been shown to lead to low SP-D levels [16] . In the present study, we assessed the potential association of missense polymorphisms of the SFTPA1, SFTPA2 and SFTPD genes as well as the resulting haplotypes, with the susceptibility to and the severity and outcome of CAP in adults. In addition, we evaluated the existence of linkage disequilibrium (LD) among these genes, and the effect of genetic variability on SP-D serum levels. We studied 682 patients and 769 controls, all of them Caucasoid Spanish adult individuals from five hospitals in Spain. Foreigners and individuals with ancestors other than Spanish were previously excluded in the selection process. The diagnosis of CAP was assumed in the presence of acute onset of signs and symptoms suggesting lower respiratory tract infection and radiographic evidence of a new pulmonary infiltrate that had no other known cause. A detailed description of the exclusion criteria and clinical definitions are shown in Methods in Additional File 1 [17] [18] [19] . The control group was composed of healthy unrelated blood donors from the same hospitals as patients. For susceptibility, a case-control study was performed. Severity and outcome were evaluated in a prospective study of CAP patients. Demographic and clinical characteristics of CAP patients included in the study are shown in Table E2 in Additional File 1. In order to analyze the effect of the SFTPD aa11 on SP-D levels in our population, protein levels were measured in serum samples from individuals in the control group by means of a Surfactant Protein D ELISA kit (Antibodyshop ® , Gentofte, Denmark). Four haplotypes of SP-A1 (6A, 6A 2 , 6A 3 and 6A 4 ) and six of SP-A2 (1A, 1A 0 , 1A 1 , 1A 2 , 1A 3 and 1A 5 ) are found frequently (>1%) in the general population [15] . On the basis of the differences in non-synonymous SNPs (SFTPA1-aa19, -aa50, -aa219, SFTPA2-aa9, -aa91, -aa223) the most frequent conventional haplotypes of these genes, except 1A and 1A 5 , can be unambiguously identified (see Table E1 in Additional File 1). However, this method does not allow for the differentiation of some of these haplotypes from those rare haplotypes (frequency equal or lower than 1%) identified with the SNPs indicated in Table E1 in Additional File 1. For comparative purposes, in our study each haplotype was denoted by the name of the most frequent haplotype for a given combination of non-synonymous SNPs. Genomic DNA was isolated from whole blood according to standard phenol-chloroform procedure or with the Magnapure DNA Isolation Kit (Roche Molecular Diagnostics, Pleasanton, CA, USA). Genotyping of polymorphisms in SFTPA1 (aa19, aa50, aa219), SFTPA2 (aa9, aa91, aa223) and SFTPD (aa11) genes was carried out using minor modifications of previously reported procedures [15, 20] . The accuracy of genotyping was confirmed by direct sequencing in an ABI Prism 310 (Applied Biosystems, Foster City, CA, USA) sequencer. Haplotypes for each individual were inferred using PHASE statistical software (version 2.1) [21] . The haplotype of SFTPA1, SFTPA2 or the haplotype encompassing SFTPA1, SFTPA2 and SFTPD was ambiguous or could not be assigned in 12 individuals, who were excluded from the study. The order used for the haplotypes nomenclature is SFTPD-SFTPA1-SFTPA2. Linkage disequilibrium (LD) was measured by means of Arlequin (version 3.11) [22] and Haploview [23] softwares in the control group. In addition, pairwise LD between haplotypes of SFTPA1 and SFTPA2 as well as with the SFTPD SNP was characterized using Arlequin 3.11. The existence of LD was considered if D' >0.4. Informed consent was obtained from the patients or their relatives. The protocol was approved by the local ethics committee of the five hospitals. All steps were performed in complete accordance to the Helsinki declaration. Bivariate and multivariate statistical analyses were performed using SPSS (version 15.0) (SPSS, Inc, Chicago, Ill, USA) and R package [24] . A detailed description of the statistical methods is shown in Methods in Additional File 1. Susceptibility to CAP related to SFTPA1, SFTPA2 and SFTPD gene variants Seven non-synonymous SNPs were genotyped across the region containing the SFTPD, SFTPA1 and SFTPA2 genes ( Table 1) . None of the SNPs showed a significant deviation from Hardy-Weinberg equilibrium in controls. Several major alleles were overrepresented in controls compared with patients, but only SFTPA1 aa50-G, SFTPA2 aa9-A and aa91-G remained significant after Bonferroni correction for multiple comparisons. A dominant effect of SFTPA2 aa9-A, and a recessive effect of SFTPA1 aa50-G and aa219-C as well as SFTPA2 aa223-C were associated with a lower risk of CAP (see Table 1 ). When haplotypes were inferred, seven different haplotypes were found for SFTPA1 and eight for SFTPA2 (see Table 2 ). All haplotypes except 6A 5 , 6A 15 , 1A 10 and 1A 13 had frequencies higher than 1% in our population. The most frequent haplotype for SFTPA1 and SFTPA2 were respectively TGC and AGC, which correspond mainly with the 6A 2 and 1A 0 haplotypes respectively. The frequencies of both haplotypes were significantly lower in patients compared to controls (P = 0.0009, OR = 0.78; 95% confidence interval (CI) 0.67 to 0.91, for SFTPA1 6A 2 . P = 0.002, OR = 0.79; 95% CI 0.68 to 0.92, for SFTPA2 1A 0 ), even when Bonferroni correction was applied. Several haplotypes were overrepresented in patients compared with controls, but only 1A 10 (P = 0.00007, OR = 6.58; 95% CI 2.24 to 26.22) remained significant after Bonferroni correction. For the observed odd-ratios, the power of the tests with a significance level of 1% were 84.16%, 79.09% and 94.04% for the haplotypes 6A 2 , 1A 0 and 1A 10 respectively. In addition, dominant and recessive models showed a significant dominant effect on CAP susceptibility for haplotypes 6A 3 , 1A 0 , 1A 7 and 1A 10 and a recessive effect for haplotype 6A 2 (see Table 2 ). Linkage disequilibrium of SFTPA1, SFTPA2 and SFTPD genes Pairwise LD (D') measured by means of Arlequin confirmed the existence of LD among several SNPs at SFTPA1 and SFTPA2, whereas SFTPD aa11 was only observed in LD with SFTPA1 aa19 (see Figure 1) . A similar pattern of LD was observed when D' was measured by means of the Haploview software (data not shown). SFTPA1 and SFTPA2 were previously found to be in LD [25, 26] . The value of LD measured as r 2 was very low for every pair of SNPs (data not shown), and none of the studied SNPs could be used as haplotypetagging SNP to infer the observed haplotypes. When pairwise LD was measured among haplotypes instead among SNPs, SFTPA1 was found to be in LD with SFTPD aa11, but only a marginal LD was found between SFTPA2 1A and SFTPD aa11 (see Table E3 in Additional File 1). Susceptibility to CAP related to haplotypes encompassing SFTPA1, SFTPA2 and SFTPD When haplotypes encompassing both SFTPA genes were studied, we observed 39 of the 64 expected haplotypes, and only 14 haplotypes had frequencies higher than 1% (data not shown). The most common SFTPA1-SFTPA2 haplotype, 6A 2 -1A 0 , was underrepresented in patients (P = 0.0005, OR = 0.77; 95% CI 0.66 to 0.90), whereas 6A 3 -1A was overrepresented (P = 0.0007, OR = 3.92; 95% CI 1.63 to 10.80) (see Table 3 ). Both differences remained significant after Bonferroni correction. For the observed odd-ratios, the powers of the tests with a significance level of 1% were 87.76% and 84.04% for the haplotypes 6A 2 -1A 0 and 6A 3 -1A respectively. On the other hand, dominant and recessive logistic regression models showed a significant dominant effect on CAP susceptibility for haplotypes 6A 3 -1A and 6A-1A 1 and a recessive effect for haplotype 6A 2 -1A 0 (see Table 3 ). We also intended to analyze whether phased variants encompassing the three genes were involved in susceptibility to CAP. Only 68 of the 128 expected haplotypes were observed, and 16 of them had a frequency over 1%. Chromosomes containing C-6A 2 -1A 0 were decreased in patients when compared with controls (P = 0.00001, OR = 0.62; 95% CI 0.50 to 0.77), a difference that remained significant after Bonferroni correction. C-6A 2 -1A 0 was also significantly associated with protection against CAP in a dominant model (see Table 3 ). A similar pattern of haplotype distribution was observed when individual as well as two-and three-gene based haplotypes were compared between pneumococcal CAP patients and healthy controls (see Table E4 in Additional File 1), though no significant differences were now observed after Bonferroni corrections. Outcome and severity of CAP patients related to genetic variants at SFTPA1, SFTPA2 and SFTPD genes When fatal outcome was analyzed, patients who died within the first 28 days showed a higher frequency of haplotypes 6A 12 , 1A 10 and 6A-1A, and a lower frequency of the major SFTPA1aa19-T and aa219-C alleles and of haplotypes 6A 3 and 6A 3 -1A 1 (see Table 4 ). Similar results were observed when 90-day mortality was analyzed (see Table 4 ). For the observed odd-ratios, the power of the tests with a significance level of 5% was 82.64% when the protective effect of 6A 3 -1A 1 on 28-day mortality was evaluated, and 81.45% and 80.79% concerning the effect of 6A 3 and 6A 3 -1A 1 on 90-day mortality respectively. Kaplan-Meier analysis ( Figure 2 ) and log-rank test (Table 4 ) also showed significantly different survival for the above mentioned alleles and haplotypes. Cox Regression for 28-day survival, adjusted for age, gender, hospital of origin and co-morbidities, was significant for haplotypes 6A 12 and 6A-1A, and it remained significant for haplotypes 6A 3 and 6A-1A when 90-day survival analysis was performed (see Table 4 ). We also analyzed Cox Regression adjusted for hospital of origin, PSI and pathogen causative of the pneumonia, and we found similar results: for 28-day Figure 1 Genomic organization, location of SNPs, and linkage disequilibrium (D') map for SFTPD, SFTPA1 and SFTPA2 genes. SNPs: Single-nucleotide polymorphisms. All the D' values higher than 0.3 were statistically significant (P < 0.05). Linkage disequilibrium was measured in the control group. survival it remained significant for haplotype 6A-1A (P = 0.029, OR = 2.45; 95% CI 1.10 to 5.46), although for 6A 12 haplotype it was not significant (P = 0.072); for 90-day survival it was significant for both 6A 3 (P = 0.038, OR = 0.52; 95% CI 0.28 to 0.96) and 6A-1A (P = 0.045, OR = 2.12; 95% CI 1.02 to 4.44) haplotypes. No effect of the SFTPD aa11 SNP was observed. Due to the high number of observed haplotypes, and because of the limited sample size in the patient groups when they were stratified on the basis of severity and outcome, the haplotypes including SFTPA1, A2 and D were not studied. The relevance of these genetic variants in the severity of CAP was also evaluated by analyzing predisposition to acute respiratory distress syndrome (ARDS) and to multiorgan dysfunction syndrome (MODS) (see Tables 5 and 6 ). The SFTPD aa11-C allele was significantly overrepresented in patients with MODS or ARDS. Haplotypes 6A and 6A-1A, were also associated with the development of ARDS, and SFTPA2 1A and 1A 10 were associated with the development of MODS. For the observed odd-ratios, the power of the association of 1A with predisposition to MODS was 89.29%. However, the number of individuals included in the analysis of outcome was relatively small and the power of the tests with a significance level of 1% was lower than 80%. These associations remained significant in multivariate analysis adjusted for age, gender, hospital of origin and co-morbidities, as well as for hospital of origin, PSI and causative microorganism (see Tables 5 and 6 ). By contrast, 6A 3 -1A 1 was associated with protection against MODS, although this difference was not significant in the multivariate analysis. In order to study whether variants at the pulmonary collectins were associated with differences of serum levels of SP-D, this protein was measured in serum from healthy controls with known genotypes. The SFTPD aa11-C SNP associated with lower SP-D serum levels (905.10 ± 68.38 ng/ml for T/T genotype, 711.04 ± 52.02 ng/ml for T/C, and 577.91 ± 96.14 ng/ml for C/C; ANOVA P = 0.017) (see Figure 3 ). This study is unique in reporting a genetic association between non-synonymous SNPs at SFTPD, SFTPA1 and SFTPA2, as well as of haplotypes encompassing these genes, with the susceptibility, severity and outcome of CAP. The major alleles of SFTPA1 aa50-G, aa219-C as well as SFTPA2 aa9-A and aa91-G or genotypes carrying these alleles were associated with protection against CAP. The frequencies of the different SNPs and haplotypes of SFTPA1, SFTPA2 and SFTPD observed in our study were similar to those previously reported in European populations [25] . SFTPA1 and SFTPA2 were reported to be in strong LD [26, 27] , and several haplotypes of these loci tend to segregate together, being 6A 2 -1A 0 the major haplotype [27] . A protective role against CAP was associated with 6A 2 , 1A 0 and 6A 2 -1A 0 in our survey but only the rare 1A 10 and 6A 3 -1A haplotypes were significantly associated with susceptibility to CAP. Similar results were observed in susceptibility to pneumococcal CAP. Several SNPs and Table 4 . [15] . † P-value for the bivariate comparison. ‡ P-value for multivariate analysis, including the variables age, gender, hospital of origin and co-morbidities. For those bivariate comparisons that resulted in nonsignificant differences, multivariate analysis were not calculated. § P-value for multivariate analysis, including the variables hospital of origin, PSI (Pneumonia Severity Index) and pathogen. haplotypes were also associated with a higher severity and poor outcome; MODS, ARDS, and mortality were selected because they represent the more severe clinical phenotypes. Particularly, 1A 10 and 6A-1A were overrepresented among patients who died at 28 or 90 days, and they also predisposed to MODS and ARDS respectively. Likewise, 6A was associated with ARDS, and 1A was associated with MODS. By contrast, 6A 3 and 6A 3 -1A 1 were underrepresented in patients who died. The SFTPD aa11-C allele was associated with the development of MODS and ARDS, but no significant effects on mortality were observed. In spite that the power of the test for some associations with outcome and severity were higher than 80% for the observed OR with a significance level of 5%, the number of individuals included in the analysis of outcome was relatively small. Consequently, associations with outcome should be interpreted with caution. Only a few studies have addressed the role of the genetic variability at SFTPA1, and SFTPA2 in infectious diseases [28] [29] [30] [31] . In bacterial infections, homozygosity for the 1A 1 haplotype was reported to be associated with meningococcal disease [30] . Noteworthy, 6A 2 -1A 0 was protective against acute otitis media (AOM) in children [32] . Haplotypes 6A 2 and 1A 0 may also be involved in protection against respiratory syncytial virus (RSV) disease [29, 33] . Considering the high difference in the frequencies with the corresponding alternative alleles and haplotypes, it is tempting to speculate that 6A 2 , 1A 0 and 6A 2 -1A 0 could have been maintained at high frequencies partly by their protective effect against respiratory infections. The 6A and 6A-1A haplotypes were found to be associated with an increased risk of wheeze and persistent cough, presumably triggered by respiratory infections or environmental contaminants, among infants at risk for asthma [27] . Regarding SP-D, the SFTPD aa11-T allele was associated with severe RSV bronchiolitis [34] , whereas the SFTPD aa11-C variant was associated with tuberculosis [30] . In sharp contrast to the potentially proinflammatory effects after PAMP recognition by collectins, mice deficient in SP-A or SP-D develop enhanced inflammatory pulmonary responses [35] [36] [37] . SP-A and SP-D play a dual role in the inflammatory response. They interact with pathogens via their CRD, and are recognized by calreticulin/CD91 on phagocytes through the N-terminal collagen domain, promoting phagocytosis and proinflammatory responses [10, 13] . By contrast, binding of the CRD to signal inhibitory regulatory protein α (SIRPα) on alveolar macrophages suppresses NF-B activation and inflammation, allowing the lung to remain in a quiescent state during periods of health [10] . A similar dual effect is observed in the promotion or inhibition of apoptosis [12] . SP-A and SP-D can also inhibit inflammation by blocking, through the CRD, Toll-like receptors 2 and 4 [38, 39] . In agreement with previous results [16] , we have observed that the SFTPD aa11-C allele associates with significantly lower SP-D serum levels than the aa11-T allele, and this effect was dose-dependent. The aa11-C/T SNP, located in the Nterminal domain, influences oligomerization of SP-D and explains a significant part of the heritability of serum SP-D levels [16, 40] . Serum from aa11-C homozygotes lack the highest molecular weight (m.w.) forms of the protein, which binds preferentially to complex microorganisms whereas the low m.w. SP-D preferentially binds LPS [16] . As a consequence of intracellular oligomerization, monomeric SP-A subunits fold into trimers, and supratrimeric assembly leads to high-order oligomers [41, 42] . The degree of supratrimeric oligomerization is important for the host defense function [14, 41, [43] [44] [45] . SP-A1 and SP-A2 differ in only four amino acids (residues 66, 73, 81 and 85) located in the collagen domain [46] . In most functions examined, recombinant human (rh) SP-A2 shows higher biological activity than SP-A1 [14, 41, [47] [48] [49] [50] . The significance and the nature of functional differences between variants at SP-A1 and SP-A2 are poorly understood [14, 49, 50] . Variants aa50 (SP-A1) and aa91 (SP-A2) are located in the collagen region. These changes may affect the oligomerization pattern and binding to receptors such as calreticulin/CD91 or the functional activity of the protein. Likewise, the variants aa219 (SP-A1) and aa223 (SP-A2) are located in the CRD, and might directly influence the binding properties to microorganisms or to surface receptors such as SIRPα or TLR4. Residue 9, and frequently residue 19, is located in the signal peptide, and it is not know whether these variants may affect the function of the protein [14, 44] . Alternatively all the missense variants could be in LD with SNPs in regulatory regions that might affect translation and RNA stability [51, 52] . Native SP-A is thought to consist of hetero-oligomers of SP-A1 and SP-A2, and properties of co-expressed SP-A1/SP-A2 are between those of SP-A1 and SP-A2 [41, 46] . However, the extent of oligomerization of SP-A, as well as the SP-A1/SP-A2 ratio, may be altered in various diseases and can vary among individuals [53, 54] . The combination of both gene products may be important for reaching a fully native conformation [41] . In fact, it was recently shown that both SP-A1 and SP-A2 are necessary for the formation of pulmonar tubular myelin [55] . Therefore, the effect of a given haplotype may be largely influenced by haplotypes at the other gene. Our results suggest that the 6A 2 to1A 0 haplotype is more protective against CAP than both 6A 2 and 1A 0 . It was previously reported that the SFTPD aa11 SNP is in LD with SFTPA1 and SFTPA2 [25] . A protective effect of the 6A 2 to 1A 0 haplotype was even higher when this haplotype co-segregates with the SFTPD aa11-C allele. Likewise, one haplotype containing 6A 2 -1A 0 and the G allele of the SFTPD aa160 SNP could be protective against severe RSV disease [29] . Haplotypes at SFTPA1 are in LD with SFTPD aa11 in our population, but only a marginal LD between SFTPA2 and SFTPD aa11 was observed. In addition, no LD between 6A 2 to A 0 and SFTPD aa11 was found in controls (D' = 0.09) or CAP patients (D' = 0.024) in our study. These findings suggest that the protective effect of the co-segregation of SFTPD aa11-C with 6A 2 to 1A 0 on CAP susceptibility may rather reflect genetic interactions. Alternatively, the SFTPD aa11 SNP may be a marker of other SNPs in LD with SFTPA1 and SFTPA2. The gene of another collecting, the mannose-binding lectin (MBL), is located at 10q11.2-q21. We have previously observed that MBL deficiency predisposes to higher severity and poor outcome in CAP [56] , and LD of the SP genes with MBL2 cannot be ruled out. Despite modern antibiotics, CAP remains a common cause of death, and the search for new therapeutic approaches has been redirected into non-antibiotic therapies [57] . SP-A levels are reduced in several pulmonary diseases [58] [59] [60] . SP-D may also be reduced in some patients with ARDS [59] . In Sftpa -/and Sftpd -/mice, intratracheally administered SP-A or SP-D can restore microbial clearance and inflammation [8, 35] . Exogenous surfactant preparation containing the hydrophobic SP-B and -C are nowadays widely used for replacement therapies in infantile RDS. In addition, intratracheal instillation of recombinant SP-C reduced mortality in patients with severe ARDS due to pneumonia or aspiration [61] . Some of the genetic variants analyzed in our survey, such as 1A 10 , although rare, may have a high impact on susceptibility, severity and outcome of CAP. Validation of our results in other populations, and a better knowledge of the functional and clinical significance of the genetic variability at SFTPA1, SFTPA2 and SFTPD could be relevant for future investigations in the use of these collectins in the treatment of respiratory infectious diseases. The surfactant proteins A1, A2 and D are key components of innate immune response and the antiinflammatory status in the lung. Genetic variability at the genes of these collectins influences susceptibility and outcome of community-acquired pneumonia. These results could be relevant for future investigations in the use of these collectins in the treatment of respiratory infectious diseases. • The SFTPA1 and SFTPA2 haplotypes 6A 2 , 1A 0 and 6A 2 to 1A 0 , and the SFTPD-SFTPA1-SFTPA2 haplotype C-6A 2 to 1A 0 are associated with a protective role against the development of Communityacquired pneumonia (CAP). • 1A 10 and 6A 3 to 1A haplotypes are associated with increased susceptibility to CAP. • Haplotypes 6A and 6A to 1A are associated with development of ARDS, while 1A and 1A 10 are associated with MODS in patients with CAP. • The variant SFTPD aa11-C leads to decreased SP-D serum levels, and predisposes to development of MODS and ARDS in patients with CAP. • Haplotypes 6A 12 , 1A 10 and 6A to 1A are overrepresented among patients who died at 28 or 90 days. By contrast, 6A 3 and 6A 3 to 1A 1 are protective against 28-day and 90-day mortality. Additional file 1: Further description of methods, definitions and statistical analysis, and Tables E1-E4. The file contains additional information on exclusion criteria and definitions of PSI, ARDS and MODS. The statistical tests used are described. The additional file also includes four tables. Table E1 defines the resulting haplotypes from SNPs combination in SFTPA1 and SFTPA2 genes. Table E2 presents demographic and clinical characteristics of CAP patients. Table E3 shows the pairwise linkage disequilibrium measure for surfactant proteins A1, A2 and D alleles. Table E4 compares haplotypes of SFTPA1, SFTPA2 and SFTPD between patients with pneumococcal CAP and controls.

Survival of Influenza A(H1N1) on Materials Found in Households: Implications for Infection Control

BACKGROUND: The majority of influenza transmission occurs in homes, schools and workplaces, where many frequently touched communal items are situated. However the importance of transmission via fomites is unclear since few data exist on the survival of virus on commonly touched surfaces. We therefore measured the viability over time of two H1N1 influenza strains applied to a variety of materials commonly found in households and workplaces. METHODOLOGY AND PRINCIPAL FINDINGS: Influenza A/PuertoRico/8/34 (PR8) or A/Cambridge/AHO4/2009 (pandemic H1N1) viruses were inoculated onto a wide range of surfaces used in home and work environments, then sampled at set times following incubation at stabilised temperature and humidity. Virus genome was measured by RT-PCR; plaque assay (for PR8) or fluorescent focus formation (for pandemic H1N1) was used to assess the survival of viable virus. CONCLUSIONS/SIGNIFICANCE: The genome of either virus could be detected on most surfaces 24 h after application with relatively little drop in copy number, with the exception of unsealed wood surfaces. In contrast, virus viability dropped much more rapidly. Live virus was recovered from most surfaces tested four hours after application and from some non-porous materials after nine hours, but had fallen below the level of detection from all surfaces at 24 h. We conclude that influenza A transmission via fomites is possible but unlikely to occur for long periods after surface contamination (unless re-inoculation occurs). In situations involving a high probability of influenza transmission, our data suggest a hierarchy of priorities for surface decontamination in the multi-surface environments of home and hospitals.

Influenza transmission is well documented in households and other residential settings [1] [2] [3] [4] . Yet the underlying mechanisms of transmission remain poorly understood and hotly debated [5, 6] . Although transmission by aerosols (particles typically ,5 mm in diameter), larger droplets and contact transmission (direct and via fomites) probably all play some role, the relative importance of each is uncertain, which has led to difficulties regarding the provision of evidence-based infection control advice for both pandemic and seasonal influenza [7] . If virus can survive for meaningful periods on surfaces and objects, or alternatively, if surfaces are frequently re-inoculated (e.g. by toddlers), then it is feasible that transmission via fomites might occur. The potential for transmission of influenza by indirect contact (i.e. via fomites) is linked to the ability of virus to survive in transmissible titres on commonly touched surfaces; however few data exist on this subject. Parker et al (1944) demonstrated improved survival of influenza viruses in the presence of human mucus [8] ; and in 1962, Buckland demonstrated experimentally that influenza virus was inactivated relatively quickly on glass, probably through desiccation [9] . In 1982, widely cited work by Bean et al showed that both influenza A and B, directly applied to stainless steel surfaces or hard plastic, could survive for 24-48 hours, and be transferred, from there to hands, for 24 hours; survival was much shorter on porous materials such as paper and cotton (8-12 hours) , with transferability to hands for only 15 minutes [10] . In contrast, Thomas et al, recently demonstrated survival of human seasonal A (H1N1) and A (H3N2) on Swiss banknotes for up to three days, increasing to up to eight days when applied with nasopharyngeal secretions from children (17 days if applied at very high concentration). Although viable virus was recovered at each of these time points, it was noted that virus load declined sharply after the first few days; no other materials were tested [11] . Other studies have detected influenza virus on fomites in homes and health and childcare facilities, using RT-PCR to establish the presence of the viral genome [12] [13] [14] . However, data obtained using this technique (even quantitatively) do not distinguish adequately between viable and non-viable virus and are therefore problematic to interpret in the context of practical infection control guidance. In another recent study, virus was detected by PCR on commonly touched household surfaces, but only one sample proved culture positive [15] . However, the time from deposition to recovery was not known, nor the extent of any cleaning undertaken. We evaluate the survival of influenza A (H1N1) viruses deliberately applied to a range of commonly touched household and workplace surfaces, using RT-PCR for genome detection and culture methods to determine viability. We conclude that RT-PCR is only useful to demonstrate the absence of virus and that on most surfaces, virus viability drops rapidly. Nevertheless, on certain non-porous surfaces, viable virus persists for several hours, rendering fomite transmission possible without re-inoculation. To test the surface survival of influenza virus, we used a variety of materials commonly encountered in the home and workplace, including a hospital setting ( Table 1) ; choice of surfaces to be tested was discussed with the Department of Health, England to ensure relevance to public health policy. These included fibrous materials such as the ubiquitous J-clothH (Associated Brands) widely used for cleaning, a silver impregnated fabric with known bacteriostatic properties (Toray Textiles Europe Ltd.) of the type sometimes encountered in hospital staff clothing to combat nosocomial bacterial infections, as well as fabric from a child's soft toy. The latter fabric was made of non-absorbent polyester and, although a porous item overall, individual fibres might perform as a nonporous surface. A variety of non-porous plastic surfaces representing objects highly likely to be touched by multiple individuals such as light switch, telephone and keyboard plastics were also tested, as well as porous and non-porous 'background' materials such as various wood surfaces, glass, Perspex/plexiglass (poly (methyl methacrylate) -a thermoplastic often used as a light or shatterresistant alternative to glass) and metals. As a control surface, we used standard laboratory polystyrene culture dishes. As viruses, we used two human H1N1 strains: the laboratory adapted A/Puerto Rico/8/34 (PR8) strain because of ready availability and robust, convenient assay systems with a wide dynamic range, and an isolate of the current 2009 pandemic virus A/Cambridge/AH04/ 2009 (AH04), as a low passage history representative of a virus likely to be encountered in the current environment. The source and disinfection method used to clean the various surfaces before testing are listed in Table 1 . Human influenza A virus PR8 (Cambridge lineage) was grown in embryonated hens' eggs and harvested at a titre of 9610 8 pfu/ ml. For inoculation of the surfaces, the virus was diluted 1:10 in 1% BSA and serum free media (Dulbecco Modified Eagles Medium, DMEM, Gibco, UK). This represented a viral titre approximating 1.5610 8 TCID50/ml, just above the upper end of titres reported for human shedding [10, 11] . Preliminary experiments established that virus survival was improved by the addition of extra protein to the suspension. We tested 0.5% or 1% BSA as well as four preparations of artificial mucus produced from pig stomach mucosa (NBS Biologicals), pig stomach mucin types II or III or bovine sub maxillary glands mucin, type I-S (all from Sigma Aldrich). 1% BSA had the largest effect on titre and duration of survival, followed by the bovine mucin (data not shown). In the interests of simplicity and reproducibility, 1% BSA was therefore used in all subsequent experiments. To test a 2009 pandemic influenza A (H1N1) virus strain on selected surfaces, a clinical isolate designated influenza A/Cambridge/AHO4/2009 (AH04) was passaged once in MDCK cells and then grown in Caco-2 cells (colorectal adenocarcinoma cells, ATCC HTB-37 TM ). The virus is a recent isolate from an Received sterile in packaging from manufacturer. Fumigation in a CLIII room using a Laycock Fumigator (Tolbest Ltd). doi:10.1371/journal.pone.0027932.t001 immunocompetent patient who was hospitalised briefly at the start of their illness, but recovered. This virus does not form discrete plaques in MDCK cells and could therefore not be titred by this method. Instead, virus stocks were quantified by qPCR for segment 7 [12] . Although this method scores viable and non-viable virus particles alike, preparations of wild type influenza A viruses generally have similar particle:PFU ratios and quantitative comparison of RT-PCR and other titration methods have shown good agreement [13, 14] . The AH04 stock contained 6.5610 8 genome copies/ml and was used at a 1:10 dilution as for PR8. For comparison, the PR8 stock had a genome titre of 1.6610 11 genome copies/ml. Mouse monoclonal AA5H (Abcam) was used to detect influenza NP by immunofluorescence. Surfaces were cut into 2 cm 2 pieces and sterilised by a variety of means depending on the surface to be tested (e.g. autoclaving, fumigation etc). Sterile surfaces were glued into sterile 6-well tissue culture dishes using cyanoacrylate adhesive (Henkel, UK). Preliminary experiments (data not shown) demonstrated that dried adhesive alone was non-inhibitory to influenza virus. Under the same conditions of temperature and humidity (ranges 17-21uC and 23-24% respectively), 10 ml volumes of virus were applied to six samples of each surface at the same time. Sampling was conducted immediately -time zero -to demonstrate recoverability. A cotton swab was moistened by dipping in 3 ml of virus transport medium (VTM, Remel, UK) and then wiped carefully in 6 different directions for 1 minute across the top of the surface. Keeping everything on ice, the swab was placed into the tube containing the residual (3 ml) volume of VTM and vortexed for 1 minute. After this, the sample was split directly into 6 eppendorf tubes and stored on dry ice prior to freezing at 270uC. The remaining samples in the plate were kept in a plastic, lidded box at constant temperature and humidity. At 4, 9, 24, 48 and 72 hrs, further samples were taken and stored. After initial experiments it was clear that the virus did not survive in detectable amounts for more than 24 hrs, therefore for the majority of the experiments only the first 4 time points (0, 4, 9, and 24 hrs) were taken. Initial experiments with PR8 virus also showed that loss of virus on the swab was not a major factor, with recovery of virus at time zero from polypropylene surfaces approaching 50% of initial titre (data not shown). The qRT-PCR assay used has been described previously [12] . In brief, primers and probes to the Matrix gene of influenza A were used to detect the presence of the virus on the surfaces. Samples from all time points were stored and then extracted. Virus genome was amplified to check that the quantity of virus deposited on the different surfaces was consistent and to determine whether any of the surfaces affected the genome over time. Plaque assays were performed as previously described in MDCK cells using Avicell overlays [15, 16] , in duplicate or where possible in triplicate. To detect AH04 virus by fluorescent focus assay, infectious material from swabs was first allowed to amplify by inoculation into 1610 6 MDCK cells and incubation for 48 h. Supernatant virus was then diluted 1:10 in serum free DMEM and 250 ml used to inoculate 1.5610 5 MDCK cells in a 24 well tissue culture plate. After virus absorption, the cells were overlaid with 1 ml serum free DMEM media containing 1 mg/ml Worthington's trypsin and 0.14% BSA and incubated overnight at 37uC. The following day they were fixed with 4% formaldehyde in PBS, permeabilised by the addition of 0.2% Triton 6100 in PBS for 5 minutes at RT and fluorescently stained with anti-NP monoclonal antibody and counterstained for DNA with 4,6-diamino-2-phenylindole (DAPI) as previously described [17] . Cells were examined blind by two people and scored semi quantitatively for the presence of infected cells using a standardised schema (2 no fluorescence seen; +/2 some fluorescence seen (,5% cells infected); +5-10% cells infected; ++ .10% cells infected). The literature indicates immunofluorescence to be at least as sensitive in general as plaque assay [13, 18, 19] . Confirming this, tests using serial dilutions of known quantities of PR8 virus, our method reliably detected 20 PFU of virus in the original sample prior to amplification and 50% of the time detected 2 PFU (data not shown). To test the surface survival of the virus genome, replicate samples of the various materials were inoculated with 10 ml samples containing 1610 6 PFU of virus and incubated for defined periods of time before sample recovery was attempted by swabbing. It was noted that the liquid was absorbed by the wooden surfaces within 5 minutes whereas a droplet could be seen on non-porous surfaces for considerably longer, although in all cases, surfaces had dried by 7 hours. Material eluted from the swabs was then titred for virus genome by quantitative RT-PCR. For both PR8 (Fig. 1A , Table 2 ) and AH04 (Table 3) viruses the results were unambiguous. On most surfaces, the viral genome persisted well, with only around a 10-100 fold drop from the initially recoverable titre after 24 h. The exceptions were unsealed wood surfaces, where both viruses lost genome titre rapidly and on pine surfaces in particular, became undetectable after a few hours. Thus in general, viral RNA survives well for at least 24 h and few surfaces had any significant 'contact effect' in immediately reducing genome titre. When PR8 surface viability was assessed by plaque assay, virus inoculated onto a control surface of a tissue culture dish could be recovered efficiently at t0, but thereafter infectivity fell away rapidly with no live virus recovered at 24 h (Table 4 ). Fitting the data to a one-phase exponential decay model (Fig. 1B ) estimated the t 1/2 of the virus under these conditions to be around 1.5 h. A similar pattern of rapid loss of infectivity was seen when the household surface samples were tested, with the difference that greater initial losses of infectivity ranging between 20-fold (telephone handset) to nearly 4000-fold (unsealed pine) were seen (Table 4 ). Nevertheless, viable virus was recovered at 4 h (but not later) from the silver-impregnated cloth, soft toy fabric and in trace quantities, from light switch material. The only material (other than the control tissue culture dish) for which even low amounts of viable virus could be detected at 9 h was stainless steel. Thus despite the persistence of the viral genome on a wide variety of household surfaces, PR8 infectivity decayed sharply, with evidence of significant contact effects from some materials; most notably unsealed pine, but also a wide variety of other porous and nonporous surfaces. To test whether these findings could be extrapolated to a currently circulating virus, we next tested the survival of AH04 virus, a 2009 pandemic isolate, on a subset of the materials. Unlike PR8, as a recent clinical isolate this virus does not grow to high titres in the laboratory and nor was a workable plaque assay available. We therefore used a fluorescent focus assay in which live virus is detected by immunofluorescent detection of the viral nucleoprotein in infected cells. To boost the sensitivity with which viable virus could be detected, infectious virus present in the swabs was first amplified by growth in MDCK cells before subsequent assay. The assay therefore provides a highly sensitive but semi quantitative measure of virus infectivity, ideally suited to working with low titre samples [13, 19] . By this measure, the AH04 virus persisted for at least 24 h on the control tissue culture dish material, although titres were evidently lower at 9 and 24 h ( Table 5) . Consistent with the results obtained with PR8 virus, all household surfaces tested showed lower persistence of infectious virus, with none providing recoverable titre at 24 h and the majority failing to produce live material at 9 h. Once again the pine surface showed very rapid inactivation of viability, with no infectivity recovered at 4 h. Thus both an historic virus isolate and an example of the recent pandemic strain fail to survive in high titres for long periods of time on a variety of household surfaces, but with significant survival over shorter time spans on certain materials. Prior to the influenza A(H1N1) pandemic of 2009-10, few data were available with regard to virus survival on different household surfaces. With a few notable exceptions [10, 11] , the majority of studies had been carried out based on RT-PCR to detect the presence of the genome [20] [21] [22] ; these shed no light on the presence or absence of viable virus. In this study we sought to provide contemporary data about virus survival on a wider range of materials found in or on household surfaces than previously described in the literature; these exemplars were chosen after discussion with UK pandemic policy makers. However, one limitation is that our study was confined to H1N1 influenza A viruses (PR8 and the 2009 pandemic virus) due to resource issues. However, we know of no evidence to suggest there are substantial differences in survival between human influenza viruses. Moreover, when we compared the survival of PR8 virus with two seasonal isolates of influenza A (A/Solomon Islands/12/5/08 (H1N1) and influenza A/Brisbane/12/5/08 (H3N2), obtained from Professor Alan Hay of the National Institute of Medical Research, Mill Hill), we saw no significant differences, with all three viruses losing plaque titre on a plastic surface with a t1/2 of around 90 minutes (data not shown). We therefore think it is reasonable to generalise from the findings here to other human strains of influenza A. Further studies, especially of influenza B are warranted however. We applied concentrations of virus (,1610 6 TCID 50 ), which were within the range of those reported in the respiratory secretions of naturally infected individuals [5, 10] . In addition, we suspended virus in 1% bovine serum albumin (BSA), reflecting our (unpublished) finding that BSA improved virus survival, and similar findings from Thomas et al [11] using mucus obtained from children. Our experiments were conducted within a narrow range of humidity and temperature conditions consistent with normal human indoor living conditions in temperate zones, and all survival assays were performed in duplicate and where possible triplicate. We used plaque assay techniques and immunofluorescence techniques for PR8 and pandemic viruses, respectively. The differing methodologies used to detect the two strains of H1N1 virus (lower titre inoculum of the pandemic AH04 virus but higher sensitivity detection method) make it difficult to directly compare the survival of the two strains, but we see little to suggest any major difference. Our data on the survival of the laboratory adapted PR8 virus indicated that viable virus was no longer recoverable in detectable amounts from 9 of 14 (64%) surfaces four hours after deposition; however, contrary to the findings of Bean et al., non-porous surfaces were not consistently more conducive to virus survival than porous ones [10] . Nevertheless, no test surfaces supported detectable virus survival beyond nine hours. Broadly similar outcomes in which infectivity tended to be lost after 4-9 hours were obtained with the recent pandemic isolate AH04. Overall, our results indicate that influenza virus does not remain viable in large quantities on most surfaces in indoor domestic conditions for more than a few hours. Our data are consistent with recent findings from a study of environmental deposition of pandemic H1N1 virus in the homes of infected patients, involving our laboratory, when almost 10% of tested surfaces yielded viable virus [15] . However, in this and similar studies in community settings where environmental samples are taken relatively infrequently and the infectious source remains present, it is not possible to establish the time elapsed since virus deposition [15, 23] . With regard to the testing of specific materials, we examined survival on a range of porous items: a children's soft toy, a silver impregnated fabric with known bacteriostatic properties of the type sometimes encountered in hospital staff clothing to combat nosocomial bacterial infections, and a branded cleaning cloth (J clothH, Associated Brands). We hypothesised that the inclusion of an antimicrobial agent, MicrobanH (Microban International Ltd) in the J cloth might inhibit viral growth. MicrobanH is based on triclosan and has been demonstrated to have anti-bacterial and anti-fungal activity; it has not however, been demonstrated or claimed to be anti-viral. Notwithstanding, in our laboratory setting, some constituent or quality of the J clothH appeared to limit virus survival to under 4 hours. The result for the silver impregnated fabric also deserves further comment. Whilst silver has been demonstrated to have bacteriostatic properties, it has not been documented to show antiviral activity. Our data would tend to suggest that it is not significantly inhibitory to influenza A. Surfaces that allowed PR8 virus to survive longest (between four and nine hours) included light switch material (polyvinyl chloride) and a computer keyboard. Interestingly these are likely to be the materials from which the most frequently touched communal household objects are made. Both PR8 and pandemic viruses survived less than four hours on all of the wood surfaces tested. This may have been due to a number of factors including porosity of the surface, oils in the wood or a potentially virucidal 'contact effect' of varnish finishes. Pine oil in particular has been demonstrated to have virucidal activity against respiratory viruses [24] . Our findings suggest they are not hospitable environments for enveloped viruses. As observed in other studies, we found that stainless steel supported the viability of influenza viruses longer than other tested metals. Metals have been demonstrated to have low levels of anti viral activity [25] [26] [27] ; and stainless steel has previously been demonstrated to support influenza virus viability for longer than that of copper [28] . Confirmation of these results raises questions about the use of stainless steel in healthcare and daycare settings in particular. In conclusion, testing two H1N1 strains of influenza A (one of which was a 2009 pandemic virus) demonstrates that in an environment that is consistent with indoor domestic settings in temperate zones, virus deposited onto the touched environment is likely to survive up to a few hours, though rarely more than nine hours, on the vast majority of surfaces. Metallic and non-metallic non-porous materials pose the greatest risk and should be targeted for frequent cleaning if situated in close proximity to patients infected with influenza virus; fortunately the latter are also more conducive to surface cleaning with a wide variety of simple cleaning agents [12] . Whilst our data suggest that the risk of virus transmission might last several hours after deposition, we generated very little data suggesting that appreciable amounts of virus survived much beyond nine hours. This probably means that frequently touched environments such as classrooms, offices and living rooms, which are then left unoccupied overnight, will not contain much viable virus on surfaces by the next morning. Nevertheless, the data still support frequent cleaning of commonly touched items and surfaces throughout the working day, particularly when symptomatic persons are present, for example in physician waiting rooms. In terms of cleaning regimens, one critically important consideration is that survival of virus in high titres for prolonged periods is not necessary for fomite transmission if surfaces are frequently re-inoculated (e.g. by toddlers). However the contribution of such indirect transmission relative to respiratory droplets directly from one person to another or relative to aerosol transmission remains unknown.

Respiratory failure presenting in H1N1 influenza with Legionnaires disease: two case reports

INTRODUCTION: Media sensationalism on the H1N1 outbreak may have influenced decisional processes and clinical diagnosis. CASE PRESENTATION: We report two cases of patients who presented in 2009 with coexisting H1N1 virus and Legionella infections: a 69-year-old Caucasian man and a 71-year-old Caucasian woman. In our cases all the signs and symptoms, including vomiting, progressive respiratory disease leading to respiratory failure, refractory hypoxemia, leukopenia, lymphopenia, thrombocytopenia, and elevated levels of creatine kinase and hepatic aminotransferases, were consistent with critical illness due to 2009 H1N1 virus infection. Other infectious disorders may mimic H1N1 viral infection especially Legionnaires' disease. Because the swine flu H1N1 pandemic occurred in Autumn in Italy, Legionnaires disease was to be highly suspected since the peak incidence usually occurs in early fall. We do think that our immediate suspicion of Legionella infection based on clinical history and X-ray abnormalities was fundamental for a successful resolution. CONCLUSION: Our two case reports suggest that patients with H1N1 should be screened for Legionella, which is not currently common practice. This is particularly important since the signs and symptoms of both infections are similar.

Media sensationalism with respect to the swine flu outbreak may have influenced decisional processes and clinical diagnosis. We report two cases of patients who present during 2009 in whom H1N1 and Legionella infection coexisted. Secondary bacterial pneumonia is recognized as one of the most common causes of death in influenza cases. Coinfection has been found in 30% of all influenza cases in persons with seasonal influenza. The pathogens most often involved are Streptococcus pneumoniae, Staphylococcus aureus, and Haemophilus influenza [1, 2] . From July 2009 through February 2010 in Italy, 2500 confirmed cases of pandemic influenza and four and a half million cases of influenza-like illnesses were reported to the sentinel surveillance system. A total of 1278 (50%) confirmed cases of H1N1 were hospitalized. Of these, 271 (21%) cases presented with pneumonia, which was attributed to bacterial coinfection in 33 cases. Of the 33 cases with pneumonia due to a bacterial coinfection, six (18%) were due to the Legionella pneumophila serogroup 1 [3] . Our first case is a 69-year-old Caucasian man with a past medical history of coronary artery disease, chronic renal insufficiency, hypertension and type 1 diabetes. Two weeks earlier, he had been exposed to a child with an upper respiratory infection. He lived in a rural area. He had no history of insect bites, but was exposed to farm animals and pond water. He had been well until nine days earlier, when dry cough, myalgias, fever (39.4°C), malaise, sore throat and nasal congestion presented. On physical examination in the Emergency Room (ER), the patient had a temperature of 39.2°C, a heart rate of 50 beats per minute and a respiratory rate of 35 breaths per minute. A buccal swab was negative for influenza A and B antigens, and no parasites were seen on a peripheral-blood smear. Acetaminophen, ketorolac, levofloxacin and normal saline were administered. After 24 hours he presented with persistent fever (39.0°C), dry cough and respiratory failure and was admitted to the intensive care unit (ICU). Vital signs were as follows: blood pressure 135/70 mm Hg; heart rate 50 beats per minute; respiratory rate 34 breaths per minute; oxygen saturation 88%, on 50% inspired oxygen. On physical examination rhonchi were detected in the lower lung fields. Repeated tests of nasopharyngeal secretions for influenza viruses, parainfluenza virus, respiratory syncytial virus, and adenovirus were negative. Testing for antibodies to toxoplasma was suggestive of past infection. Cultures of specimens of blood, urine, and sputum were sterile. Polymerase chain reaction determination of buccal swab for H1N1 influenza A virus was positive. Urinary tests for Legionella antigens were positive. On admission main laboratory examations were as follows: white blood cell count (WBC) was 9.8 K/mL (87% neutrophils, 4% lymphocytes); C-reactive protein 205 mg/L; serum sodium 132 mEq/L; serum phosphorus 2.3 mg/dL; serum glutamate pyruvate transaminase (SGPT) 175 IU/L; serum oxaloacetate transaminase (SGOT) 184 IU/L; serum ferritin 4100 ng/mL; creatinine phosphokinase (CPK) 241 IU/L. Winthrop scale score was > 15. Chest radiograph showed low lung volumes, with patchy air-space disease consistent with multifocal pneumonia (Figure 1 ). Intravenous azithromycin (500 mg twice daily), levofloxacin (500 mg twice daily) and oral oseltamivir (150 mg twice daily) were administered. Within 18 hours after arrival, tachypnea and hypoxemia (PaO 2 : 58 mm Hg, while breathing 50% oxygen) increased further requiring intubation and mechanical ventilation. Hypotension and renal failure developed; methylprednisolone and vasopressors were administered. On the third day reverse transcriptasepolymerase chain reaction (RT-PCR) on a broncoalveolar lavage specimen was still positive for H1N1 influenza infection. Legionella antigens were also confirmed positive. On the fifth day he was extubated and non-invasive ventilation was started. He was discharged from the ICU on day 21. Our second case is a 71-year-old Caucasian woman with a past medical history significant for hypertension, type 1 diabetes and chronic hepatitis C. She reported an eight day history of dry cough and fever (39.2°C) associated with sore throat and nasal congestion. She lived in a peripheral urban area. On emergency room examination, her Glasgow Coma Score (GCS) was 12 (E = 3 V = 4 M = 5), temperature was 39.6°C, blood pressure was 80/40 mm Hg, heart rate was 54 beats per minute, respiratory rate was 40 breaths per minute and oxygen saturation was 88% on 50% inspired oxygen. Her chest radiograph was consistent with multifocal pneumonia (Figure 1 ). There were rhonchi in the left and right lung fields. A rapid test of a specimen from a buccal swab was negative for influenza A and B antigens, and no parasites were seen on a peripheral-blood smear. Acetaminophen, ketorolac, levofloxacin and normal saline were administered. After one hour she was admitted to the ICU and due to worsening hypoxemia (PaO2 48 mm Hg, while breathing 50% inspired oxygen), neurological impairment with a GCS of 10 (E = 2 V = 3 M = 5) and hemodynamic instability (blood pressure 80/40 mmHg, heart rate 46 beats per minute), she was intubated and mechanically ventilated. RT-PCR on a broncoalveolar lavage specimen was positive for H1N1 influenza infection. On admission main laboratory examination results were as follows: white blood cell (WBC) count was 5.8 Iannuzzi et al. Journal of Medical Case Reports 2011, 5:520 http://www.jmedicalcasereports.com/content/5/1/520 K/Ml (77% neutrophils; 5% lymphocites); C-reactive protein was 312 mg/L; serum sodium was 129 mEq/L; serum phosphorus was 2.5 mg/dL; serum SGPT were 215 UI/L; serum SGOT were 220 UI/L; serum ferritin was 5280 ng/mL; CPK was 445 IU/L. Her Winthrop scale score was > 15. Intravenous levofloxacin (500 mg twice daily) and oral oseltamivir (150 mg twice daily) were administered. On the second day, hypoxemia, hypotension and renal failure developed; norepinephrine was administered after fluid challenge. For persistent hypoxemia she was ventilated in the prone-supine position for 12 hour intervals daily. Tests of the urine for legionella antigens were positive. Azithromycin (500 mg twice daily) was added her treatment. On the ninth day she underwent percutaneous tracheostomy. She was discharged from the ICU on day 35. During the Spring of 2009, a novel influenza A (H1N1) virus of swine origin emerged to cause infections in humans in North America [4] . The pandemic was carefully followed by the media but with a touch of sensationalism that caused a widespread a sense of fear in the population. Health care systems and physicians were suddenly in the spotlight. In our cases all the signs and symptoms (including respiratory failure, refractory hypoxemia, leukopenia, lymphopenia, thrombocytopenia, and elevated levels of creatine kinase and hepatic aminotransferases) were consistent with critical illness due to infection with the 2009 H1N1 virus [1, 4, 5] . Other infectious disorders may mimic H1N1 viral infection especially Legionnaires' disease. Because the swine flu H1N1 pandemic occurred in Autumn in Italy, Legionnaires' disease was to be highly suspected since its peak incidence usually occurs in early fall. Initial attempts to diagnose H1N1 infection using immunochromatography relied on test kits developed for seasonal influenza A and B viruses, many of which proved significantly less sensitive to H1N1. Hence, tests with monoclonal antibodies that react with H1N1 but not seasonal influenza A (H1N1 and H3N2) or B viruses were developed. Recognizing viral hemagglutinin and nucleoprotein, specifically allows the detection of H1N1 virus in nasal wash fluid or nasopharyngeal fluid from patients with influenza-like illnesses. Early and rapid diagnosis of H1N1-related respiratory insufficiency needs rapid screening during a pandemic but clinicians cannot rely only on the buccal swab test and need to rule out false positive and negative cases by RT-PCR on oral/nasal fluids or bronchoalveolar lavage specimens. In case one the visit by a child with an upper respiratory infection five days before the onset of illness in the patient represents also a plausible exposure to the 2009 H1N1 virus; in case two the lack of positive anamnesis for other suspicions together with the finding of positive specimens for H1N1 influenza A infection could have caused us arrive at a fashionable diagnosis and stopped us from further investigations. We do think that the immediate suspicion of Legionella infection based on clinical history, X-ray abnormalities and Winthrop University Hospital Infectious Disease Division's diagnostic weighted point system scale (Table 1) were fundamental for a successful resolution [6, 7] . Almost all patients affected by pandemic H1N1 infections admitted to an ICU because of lung involvement receive empiric antibiotic therapy. However, preliminary clinical data have failed to demonstrate a consistent role of bacterial co-infection suggesting that severe pulmonary damage occurs as a result of viral pneumonia [1, 11] . A recent autopsy study revealed evidence of concurrent bacterial infection in 29% of cases [8] . In 45% of these the pathogen was S. pneumoniae. These findings confirm the results of previous studies of autopsy specimens showing that most deaths attributed to influenza A virus occurred concurrently with bacterial pneumonia [9] . On the other hand, they highlight the importance of treating influenza patients with both empiric antibacterial therapy and antiviral medications. According to our experience we believe that zoonotic infections had to be ruled out. The lack of known contact with animals in an immunocompetent host appears to rule out zoonotic infections, such as Coxiella burnetii. One of the patients had worked near water ponds, which could have been contaminated by animal urine, but he did not have pulmonary hemorrhage so leptospirosis seems unlikely. Without exposure to birds, Chlamydia psittaci is unlikely. Community-acquired pneumonia (Streptococcus pneumoniae, Haemophilus influenzae, S. pyogenes, or Staphylococcus aureus) can cause severe pulmonary disease, especially in patients with antecedent influenza. These pathogens should have responded to the broadspectrum antimicrobial therapy; therefore, they are unlikely to have been the sole cause of illness. Atypical bacterial pathogens such as Legionella pneumophila cause multifocal pneumonia but usually do not cause upper respiratory tract symptoms [1, 2] . Anaplasmosis could result in a lower respiratory tract disease but fulminant disease is rare and clinical improvement should have occurred with levofloxacin treatment [10] . Radiographic abnormalities also needed to be ruled out. Many critically ill patients have radiographic findings of viral pneumonitis, with bilateral interstitial and alveolar infiltrates. Multifocal and patchy abnormalities as seen in these patients have been reported in cases of 2009 H1N1 influenza A virus infection but do not completely rule out invasive bacterial infection [2] . Ground glass opacity and cavitary lobar opacity should focus attention on Legionnaire's disease [11] . Another potential contributory factor that needed to be ruled out was viral infection of the respiratory tract. Infection with adenovirus or influenza virus must be considered. Adenovirus type 14 is the most likely cause of severe viral pneumonia in adults. Radiographic findings may include lobar infiltrates, although these are more characteristic of bacterial pneumonia [12, 13] . The fact that bacterial infections should have responded to levofloxacin argues against the fact that a secondary bacterial pneumonia superimposed with influenza A or B causing severe pulmonary disease. The lack of recent travel in H5N1 (bird flu) endemic areas or exposure to sick or dead poultry argue against H5N1 influenza (bird flu) [14] . Our two case reports suggest that patients with H1N1 should be screened for Legionella, which is not currently common practice. This is particularly important since the signs and symptoms of both infections are similar. Doctors should never be dazzled by contingency and media sensationalism in decision making. With prompt identification of the bacterial etiology of pneumonia, appropriate treatment can be started with both antibacterial therapy and antiviral medications. The length of hospital stay and the mortality of both pandemic and seasonal influenza can be reduced. Written informed consent was obtained from both patients for publication of this case report and any ccompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal. Authors' contributions MI collected patient data regarding the ER and ICU and was a major contributor in writing the manuscript. OP was responsible for buccal swab and bronchoalveolar lavage specimens and for RT-PCR processing. FR interpreted radiological findings and provided the radiological differential diagnosis. GS provided a major contribution in data analysis and interpretation. RT provided a major contribution in data analysis and interpretation. EDR collected data regarding ER and ICU, contributed to the analysis and interpretation, and was a major contributor in writing the manuscript. All authors read and approved the final manuscript.

Agricultural intensification, priming for persistence and the emergence of Nipah virus: a lethal bat-borne zoonosis

Emerging zoonoses threaten global health, yet the processes by which they emerge are complex and poorly understood. Nipah virus (NiV) is an important threat owing to its broad host and geographical range, high case fatality, potential for human-to-human transmission and lack of effective prevention or therapies. Here, we investigate the origin of the first identified outbreak of NiV encephalitis in Malaysia and Singapore. We analyse data on livestock production from the index site (a commercial pig farm in Malaysia) prior to and during the outbreak, on Malaysian agricultural production, and from surveys of NiV's wildlife reservoir (flying foxes). Our analyses suggest that repeated introduction of NiV from wildlife changed infection dynamics in pigs. Initial viral introduction produced an explosive epizootic that drove itself to extinction but primed the population for enzootic persistence upon reintroduction of the virus. The resultant within-farm persistence permitted regional spread and increased the number of human infections. This study refutes an earlier hypothesis that anomalous El Niño Southern Oscillation-related climatic conditions drove emergence and suggests that priming for persistence drove the emergence of a novel zoonotic pathogen. Thus, we provide empirical evidence for a causative mechanism previously proposed as a precursor to widespread infection with H5N1 avian influenza and other emerging pathogens.

Preventing and controlling emerging zoonoses require identification of the processes that drive cross-species pathogen transmission [1] . Agricultural intensification has been proposed as a major underlying cause of pathogen emergence from wildlife and domestic animal populations into human populations [2, 3] ; however, the precise mechanisms by which this occurs have rarely been demonstrated. Specific agricultural practices may increase frequency of cross-species pathogen transmission, setting the stage for persistence of new pathogens to occur. Pathogen introduction into a partially immune population-such as a population where some individuals are vaccinated or were previously infected-can result in longer and potentially larger epidemics than introduction into a naive population [4] . We refer to this phenomenon as immunity-based, population-level 'priming' for persistence. The potential of such priming to drive zoonotic emergence has been demonstrated theoretically in the general case [4] , as a possible precursor to measles emergence [5] and as a mechanism that could facilitate widespread emergence of H5N1 avian influenza in poultry [6] . Here, we present evidence that this process is not only theoretically possible but is likely to have played a key role in the first known outbreak of Nipah virus (NiV) encephalitis, and therefore for the emergence of a lethal zoonosis. NiV is a paramyxovirus that emerged in people in Malaysia in 1998 [7 -9] . Serology, virus isolation and polymerase chain reaction detection indicate that NiV is maintained in Pteropus spp. fruit bats (flying foxes), including P. vampyrus and P. hypomelanus in peninsular Malaysia [10] [11] [12] . Transmission from flying foxes to pigs is thought to occur via saliva on fomites (discarded fruit pulp) or via faecal or urine contamination of pigsties [11] . Pigs act as amplifier hosts, enabling infection of humans via droplet transmission during respiratory infections [13] . The risk of direct transmission from bats to humans in Malaysia is believed to be low [14] . Between September 1998 and April 1999, NiV caused 246 reported cases of febrile encephalitis in humans in peninsular Malaysia and Singapore [7, 8] and an epidemic of respiratory and neurological disease in commercially farmed pigs [9, 15] . Initial human cases were seen in Tambun village and surrounding areas in Perak State, followed by a large epidemic in the southern states of Negeri Sembilan and Selangor and several cases in Singapore (figure 1a). The majority of these cases occurred in pig farmers and abattoir workers [17] . During the course of the outbreak investigation, nine cases of NiV encephalitis were retrospectively diagnosed in the Tambun area with onset dates between January 1997 and May 1998, five of which were associated with the index farm of the 1998 -1999 outbreak in Perak (figure 1a). Six of the nine cases are considered confirmed cases on the basis of serological tests performed in 1999 [16] ; the remaining three are probable cases. Several mechanisms have been previously proposed for the appearance of NiV in pig and human populations in 1998. One frequently cited explanation is that flying foxes harbouring NiV were new to the region where transmission to pigs occurred, driven there by climatic anomalies. In late 1997, an El Niñ o Southern Oscillation-related drought and burning of forested land in Sumatra and Kalimantan (Indonesia) created an atmospheric haze across Sumatra and peninsular Malaysia [18] . One of us (K.B.C., see group author list) hypothesized that this atmospheric haze caused atypical flying fox immigration from Sumatra across the Straits of Malacca into peninsular Malaysia and northward towards Perak [18] . This hypothesized movement of flying foxes into an area with large pig farms potentially outside the species' normal range was proposed to have precipitated transmission of the virus from bats to pigs. However, a retrospective diagnosis of human NiV encephalitis cases on the index farm in early 1997 indicates that bat-to-pig transmission did not result from these specific environmental conditions. In particular, seven of these cases occurred prior to the rise in airborne particulate matter that is diagnostic of the haze event, which peaked in September 1997 [18] . The recognition of cases prior to the haze event refutes the hypothesis that this event drove initial cross-species transmission (table 1) . Furthermore, at the time of the major outbreak in pigs and humans, NiV antibodies were found to be widespread in flying fox colonies within peninsular Malaysia [12] , suggesting that the outbreak was not the result of a (recent) point source introduction of the virus into the bat population, as previously suggested [18] . Since its discovery in 1999, NiV has been identified as the cause of human neurological and respiratory disease in India and Bangladesh [23, 24] , and molecular evidence suggests a wide distribution of henipaviruses in the reservoir hosts [25] [26] [27] [28] [29] . Despite the continued threat that NiV and related viruses represent to global health, no detailed studies have combined data on pig populations, bat colonies and human cases to examine the mechanisms that drove the first known and largest outbreak. This paper describes an interdisciplinary approach that synthesizes all available data to improve and update our understanding of the process of NiV emergence in Malaysia. The study has three main branches, all of which build on earlier work. First, Chua et al. [18] identified the presence of fruit trees on the index farm as a plausible link between flying foxes and pigs that could have precipitated introduction of the virus. We place this finding in a broader context by examining spatial and temporal patterns of agricultural production in peninsular Malaysia prior to and at the time of the outbreak. We also investigate, in detail, the history and status of fruit crop production on the index farm. Second, our preliminary models suggested that repeated introduction could plausibly lead to persistence of the virus in a generic large pig population [30] . In this study, we use detailed information on management practices and production patterns at the index farm to parametrize models of this population. We use these models to examine whether repeated introduction was necessary and sufficient to allow the virus to persist on the farm. We then compare model predictions of piglet mortality patterns with data from production records to assess whether repeated introduction led to persistence of the virus in this population. Third, Johara et al. [12] identified Pteropus spp. as likely reservoirs of NiV infection. Here we examine the distribution of Pteropus bats in peninsular Malaysia, present data on NiV serology from all known roosts and examine the opportunities for transmission of the virus from bats to pigs by placing these findings in the context of large-scale agricultural production patterns. Finally, we bring together the three branches of the study to describe the events that led to NiV emergence in Malaysia and Singapore and discuss the causal nature of different contributing factors. 2.1.1. Large-scale patterns. Both pig and mango (Mangifera indica) production tripled in peninsular Malaysia between the early 1970s and the late 1990s (figure 1c). The intensification of production of pigs and mangoes was loosely correlated during this time period. The marked decline in both end-of-year standing pig population (SPP) and annual mango production between 1998 and 1999 indicates that this correlation may reflect widespread dual use of agricultural land to produce both pigs and mangoes, with the decline in commercial mango harvest reflecting the widespread abandonment and culling of infected pig farms that occurred in the first half of 1999 (following the discovery of NiV). Table 1 . Proposed drivers of emergence. Hill outlined nine criteria to be used to assess causality for epidemiological outcomes with complex origins [19] . A useful discussion of the implementation of these criteria in disease ecology is provided by Plowright et al. [20] . This figure 1a ; [18] a Temporality is the only criterion that is logically necessary (in bold). Nipah virus emergence J. R. C. Pulliam et al. 93 30 000 (approx. the same size as at the time the farm was culled in 1999). Mangoes, jackfruit (Artocarpus heteropyllus) and durian (Durio sp.) were grown on the index farm, while other farms in the Tambun area grew primarily pomelos (Citrus maxima), which are not eaten by flying foxes. As Chua et al. [18] identified, several large mango trees were planted directly adjacent to pig enclosures. Approximately 400 mango trees were planted on the farm in 1983, and the trees began producing fruit in 1987. The location of these trees, in relation to the areas where pigs were kept, is shown in the electronic supplementary material, which also includes additional details of the timeline of pig and fruit production on the index farm. 2.2.1. Dynamic models. We analysed pig production data from the index farm and developed an agestructured model of NiV dynamics within the farm population. Our model shows that multiple transmission events from bats to pigs are the best explanation for the observed pattern of human cases and pig mortality. As piglets aged in the index farm, they were moved from the breeding sections into the growing section, and finally into the finishing section for the last six weeks before they were sent to slaughter. When the virus was initially introduced into the pig population, individuals entering the growing section were immunologically naive. As the virus spread throughout the farm, the proportion of pigs entering this section with active immunity increased to the point where too few susceptibles remained for the chain of transmission to continue, and the virus could not be maintained within the farm population (figure 2a,c). The first, rapid epizootic of NiV in the pig population on the index farm thus would have produced highly localized human infections (figure 1a) and the virus most probably went extinct in the pig population at this time (figures 2a,c and 3). Our models suggest that, when NiV was reintroduced to the farm, it circulated enzootically until the farm was depopulated in March 1999 (figures 2 and 3). This finding is consistent with the timing of human cases associated with the index farm and with the high antibody prevalence detected in sows and piglets in Breeding section 2 upon depopulation of the index farm [31] . The change in dynamics between initial and subsequent introductions of the virus on the index farm results from the presence of acquired immunity in the sow population and conferred immunity in the young pig population. The age structure within the growing section, along with its rapid turnover and large population size, allowed the virus to circulate enzootically. On reintroduction of the virus into the growing section, the population had a different immunological profile from the time of initial introduction. Many young pigs moving from the breeding sections into the growing section now carried maternally derived passive immunity rather than active immunity obtained from exposure to the virus (figure 2b). We estimate that maternal antibodies are lost at approximately 14 weeks of age (electronic supplementary material). Pigs born with maternal antibodies would therefore become susceptible to the virus roughly four weeks after entering the growing section, where they would remain for another six weeks. These dynamics produce a steady inflow of susceptible individuals, which is sufficient to maintain the virus over a period of 2 years or more ( figure 2b,d) , a period similar to that observed between the index case for NiV in Malaysia in 1997 and the onset of the large-scale outbreak. The population size and turnover rate of the finishing section suggest that it may also have permitted enzootic circulation of the virus (electronic supplementary material). Stochastic simulations demonstrate that priming for persistence on the index farm is the most likely scenario consistent with pig and human epidemiology for R 0 values .2.5 (up to R 0 ¼ 30, the highest value examined), and simulations with an R 0 value in this range best reproduce the very high seroprevalence (.95% for sows and . 90% for piglets) detected in Breeding section 2 at the time of depopulation [31] (electronic supplementary material). Although no direct prevalence data exist for NiV in the pig population on the index farm, the presence of the virus in the breeding sections can be inferred from production records kept by the farm's managers (electronic supplementary material) and from reported serological findings [31] . These data are consistent with the scenario predicted by the model and with the timing of human cases: there was a period of approximately five to six months between the first introduction of the virus into the pig population and subsequent, smaller peaks in piglet mortality (figure 1b), when the virus was probably reintroduced into the farm-probably from the flying fox reservoir, or possibly from other pig farms in the area. The virus appears to have spread into each of the three breeding sections between December 1996 and February 1997, with strong evidence of infection (unusually high piglet mortality) in at least two of the breeding sections (Breeding Nucleus and Breeding section 2) during January 1997, and significant mortality on a number of subsequent occasions, following an interim period in which piglet mortality returned to baseline levels (figure 1b and electronic supplementary material). 2.3.1. Bat distribution. We conducted a countrywide survey of the distribution and infection status of the two Pteropus species in peninsular Malaysia. We located 14 active P. vampyrus roost sites across peninsular Malaysia, four P. hypomenlanus roosts on islands surrounding the peninsula and one mixed roost near Langkawi Island off the northwest coast. We estimated the current distribution of Pteropus spp. bats in peninsular Malaysia using combined results of wildlife surveys and analysis of hunter licensing data (figure 4 and electronic supplementary material). The overall distribution of flying foxes in peninsular Malaysia is similar to that observed in a large-scale survey of flying fox populations conducted in 1999 [32] , although there are fewer permanent roost sites. Satellite telemetry studies show that the bats are highly mobile, moving between Indonesia, Malaysia, Singapore and Thailand [33] . We found that flying foxes are consistently present throughout Perak State, including near the index farm, where we located two seasonal P. vampyrus roosts: one in Lenggong, Perak, ,50 km from the index farm, where we observed a maximum population of approximately 2500 bats, and another near Tambun, Perak, within 2 km of the index farm, which had been previously reported in 1999 [32] and where we observed bats in 2003 -2005 [33] . The distance between these roosts and the index farm is within the bats' nightly foraging range. Flying fox roosts were not found in the more densely populated pig-farming regions of southern Selangor and Negeri Sembilan [32, 33] (figure 4). NiV emergence in Malaysia occurred in two phases, and our current understanding of the processes leading to its emergence is outlined in figure 5 . Phase I of emergence, the occurrence of human cases, required that only two criteria be met: circulation of the virus in a local bat population and the existence of a pathway for transmission from Pteropus bats to pigs. In phase I, early human cases were directly linked to the transmission of virus from bats to pigs, with all identified cases occurring in the immediate vicinity of the index farm where bat-to-pig transmission ignited rapid pig-to-pig transmission and produced a tight cluster of human cases on the index farm and an adjacent property. Phase II of emergence, incident human cases outside the area of transmission from flying foxes, required that the criteria for phase I be met but also mandated the existence of a pathway for transmission to humans that had become unlinked from the wildlife reservoir. In principle, several scenarios could have produced such a pathway, as illustrated in figure 5 ; however, the evidence amassed here indicates that the causal pathway realized (highlighted in dark blue). Alternative pathways that would have been sufficient to produce both phases of emergence-had they occurred-are indicated by dashed arrows. Specifically, introduction of Nipah virus into an area with extremely dense pigfarming activity (such as Port Dickson in Negeri Sembilan or Sabang Perai Utara in Pulau Penang) probably could have resulted in viral persistence without priming and reintroduction. However, these areas lacked factors further up the causal pathway: the absence of Pteropus bats in these areas prevented viral introduction prior to the establishment of an alternative source of infection (e.g. persistent viral circulation on the index farm in Kinta, Perak). Similarly, if Nipah virus in Malaysia had resulted in human-to-human transmission, an alternative causal pathway could have produced incident human cases outside the area of transmission from flying foxes, as was seen in the Faridpur outbreak in Bangladesh in 2004 [34] , but there is no evidence that such human-to-human transmission occurred in Malaysia [35] . The causal pathway that was realized in Malaysia most probably involved (i) the creation of a pathway for transmission from bats to pigs via agricultural intensification (i.e. dual-use agriculture, or the practice of planting fruit trees on land used for livestock production, and increased fruit production through time) and (ii) repeated introduction of the virus into a high-turnover commercial pig population in Perak that led to viral persistence and set the stage for phase II of the emergence process. FF, flying fox. some years after the maturation of the mango trees owing to chance events within the system, e.g. fruit bat population dynamics, migratory behaviour or the dynamics of NiV in bat populations. This is supported by our own work on Hendra virus which suggests that, when the virus is actively transmitted within a bat colony, there is a heightened chance of repeated spillover, but circulation of the virus in bats is unlikely at any given place and time [36] . On the other hand, we cannot rule out the possibility that specific local conditions in late 1996 and early to mid-1997 caused increased viral shedding among bats in the area surrounding the index farm, thereby increasing the risk of cross-species transmission to pigs. Given the high mobility of flying foxes and the seasonal fluctuation in colony sizes in peninsular Malaysia [33] , NiV is probably transmitted regularly between roost sites and throughout the region. Such dynamics explain the ubiquity of NiV antibodies, and our data suggest that flying foxes and possibly NiV were regularly present near the index farm in Perak prior to the 1998 -1999 outbreak. Thus, we propose that it is plausible that flying foxes repeatedly introduced the virus onto the index farm in 1997. Our analyses have shown that the initial introduction on the index farm was most probably insufficient to induce persistent circulation of the virus. Furthermore, the lack of human cases on other farms between the time that this epizootic would have burned out and the reappearance of human cases associated with the index farm strongly suggests that the virus was not circulating on farms in the surrounding area in the interim. The short duration of this initial epizootic appears to have provided an insufficient window of time for transportation of pigs between farms to spread the infection. On the other hand, reintroduction of the virus into the 'primed' pig population on the index farm provided a substantially longer window of opportunity for spread. Our model's results indicate that reintroduction of the virus at the time of the reappearance of human cases on the index farm would have permitted enzootic circulation of the virus until the farm was culled in 1999. During this time, the infection spread to other farms in the Tambun area, probably via movement of infected pigs from the index farm, which supplied gilts and piglets to smaller operations in the vicinity. Subsequent pig movement (including 'fire sales' of pigs in reaction to the cluster of human cases in September-November 1998) allowed the virus to spread throughout Perak and eventually south to Negeri Sembilan and Selangor [15] . Because there was no evidence of transmission between humans [35] , the pattern of human infection necessarily followed the spread of the virus in pigs, explaining why human cases were much more widespread in 1998 -1999 than in 1997. In addition, the high density of pig farms in many parts of the south (electronic supplementary material) may have allowed respiratory transmission of the virus between adjacent farms without physical movement of infected pigs, contributing to the rapid spread of infection and high number of human cases in the south. On the other hand, these areas were buffered from phase I emergence via bat-to-pig transmission by the absence of flying foxes in the area [33] ( figure 4) and, possibly, a lesser degree of overlap between fruit and pig production (electronic supplementary material). We have identified two synergistic component causes [37] that precipitated each phase of emergence ( figure 5) . The existence of a pathway for transmission from flying foxes to pigs was a necessary complement to viral circulation in the bat population to produce phase I of emergence, and viral persistence in the pig population on the index farm created an infection source outside the flying fox reservoir that, combined with transportation of pigs, was sufficient to produce phase II of emergence. Each of these component causes was brought about by a phenomenon that-while not strictly necessary-did ultimately permit emergence in this context. We refer to these factors as 'drivers' of emergence, and table 1 outlines the evidence for interpreting each of these drivers as a causal factor. Agricultural intensification (namely, dual use of agricultural land and increases in production) resulted in the direct overlap of mango production and livestock rearing and therefore produced a pathway for a virus circulating in flying foxes to infect an intensively managed commercial pig population, driving phase I of emergence. Because NiV had a very low probability of persisting on the index farm without reintroduction (figure 3), priming for persistence was necessary to bring about enzootic circulation in this context and appears to have been sufficient to do so, driving phase II of emergence. This study illustrates how a broad, interdisciplinary approach to the study of emerging zoonoses focused on data from specific emergence events can illuminate emergence processes. We have demonstrated the specific role of agricultural expansion, particularly dual use of agricultural land and intensified pig production practices, in NiV emergence. A role for long-term changes in species interactions has previously been hypothesized for numerous host-jumping viral species including various primate retroviruses [38, 39] , SARS [40] and avian influenza viruses, and our evidence suggests that cross-species transmission of NiV most probably resulted from the expanded interface between wild animal reservoirs and human or domestic animal populations, although it is possible that short-term ecological conditions were also a contributing factor. Our findings have important implications for control of NiV in commercial pig farms, prevention of widespread epidemics in livestock and our general understanding of zoonotic disease emergence. First, although restrictions on planting fruit trees near pigsties appear to have prevented introduction of the virus into pig populations in Malaysia since 1999, there is a possibility of reintroduction both there and in other countries where flying foxes are found. Regular surveillance of pigs sent to markets and abattoirs in areas where pig farming overlaps flying fox distributions, along with hospital-based surveillance of encephalitis cases, may permit identification of initial NiV introductions and allow for early intervention and prevention of subsequent spread. Second, widespread prophylactic vaccination of pig populations is likely to be cost-prohibitive because of the rapid turnover of commercial pig populations. In addition, if vaccine coverage were not kept at a sufficient level, the dynamics of the virus when introduced into the population would resemble those following reintroduction on the index farm, inducing long-term persistence and increasing the risk of phase II emergence (electronic supplementary material). Finally, an important feature of the NiV outbreak in Malaysia and Singapore was the two-phase process by which emergence occurred, which implies that there was a missed opportunity for earlier recognition of this novel aetiological agent and potential intervention. Zoonotic pathogens often go unnoticed during the initial stages of 'viral chatter'-that is, repeated introductions that cause only a small number of cases [41] -because they are not yet easily transmissible in humans (e.g. simian retroviruses), are initially asymptomatic (e.g. HIV), are clinically similar to other diseases (e.g. NiV in Malaysia, which was originally misdiagnosed as Japanese encephalitis despite its epidemiological distinctiveness) and/or occur in areas with poor surveillance and diagnostic testing (e.g. NiV in Bangladesh and India). Other pathogens are noticed relatively quickly because they cause dramatic disease (e.g. Ebola haemorrhagic fever); however, even in these situations, we cannot rule out the possibility of previous undocumented cases, and the discovery of historical cases after the identification of an emerging pathogen is common (NiV [16] , SARS [42] , HIV [38] and H5N1 influenza [43] ). These patterns support previous calls for increasing targeted global surveillance and pathogen discovery in atypical disease outbreaks [44, 45] . As the Malaysian NiV outbreak highlights, surveillance of livestock and livestock workers in regions of high wildlife biodiversity would improve the chances that viral chatter of wildlife origin is detected prior to a widespread epidemic in livestock or people. This approach may be critical for identifying new outbreaks of NiV, which has recently demonstrated a potential to produce propagated outbreaks in Bangladesh, where short chains of human-to-human transmission have occurred [34] . It may also provide a strategy for the identification of unknown henipaviruses, and other novel agents, prior to epidemic or pandemic emergence. In order to better understand the history, layout and daily operation of the index farm, J.R.C.P. conducted field interviews of farm workers, including two private veterinarians who oversaw the health of the animals, and the orchard manager, as well as government veterinarians who oversaw and participated in the depopulation campaign. 4.2.1. Dynamic models. Population dynamic models of production dynamics on the index farm were developed and parametrized from descriptions of farm-management practices provided during interviews with farm workers and detailed production records from 1 January 1995 to 31 October 1996. Models of infection dynamics were then constructed by building on these baseline population models. Dynamics were explored using a deterministic ordinary differential equation (ODE) model with a cutoff value of one infected individual (calculated as the sum of individuals in all exposed and infectious classes), below which the infection was considered to have gone extinct. A stochastic individual-based model (IBM) was then used to confirm that the qualitative dynamics observed in the ODE model were robust to the incorporation of empirically derived waiting time distributions and other factors that could not be represented in the ODE framework. The stochastic model was run for a wide range of parameter combinations to assess the probability of extinction following initial and subsequent introductions under different scenarios and to assess the range of transmission parameters consistent with observed seroprevalence levels in Breeding section 2. Detailed descriptions of the ODE model and the IBM are given in the electronic supplementary material. Statistical analyses of production records were developed to detect evidence of the virus in the breeding sections. A baseline model for pre-weaning piglet mortality was formulated for each breeding section on the basis of production records from 1 January 1995 through 31 October 1996, and candidate models were compared by Akaike information criterion. For all litters born after this period, a piglet mortality index (PMI) was calculated that indicates the extent to which the observed mortality deviates from the expected mortality for a litter of the same size during the baseline period. Litter-level data are shown in the electronic supplementary material. For the sake of interpreting large-scale patterns over time, litters were binned according to farrowing date. We established a baseline for variation across bins in the average PMI based on the period from 1 January 1995 to 31 October 1996. In figure 1b, the binned PMI values are scaled in terms of the binned PMI percentile (relative to variation within the parametrization period). Absolute values and further details of the analysis are shown in the electronic supplementary material. We collaborated with the federal wildlife department and enlisted a network of sport hunters as informants to locate and monitor flying fox roost locations throughout peninsular Malaysia. A full description of census techniques (and additional findings) has been published elsewhere [33] . We also developed two indices of flying fox density based on different expectations for how hunting practices reflect bat distribution. These indices and their underlying assumptions are described in the electronic supplementary material. Flying foxes were non-randomly sampled using mist nets and anaesthetized. Three millilitres of blood was collected from the brachial vein and stored for 24 h at 48C to allow for serum separation. The separated serum was then stored at 2208C until use. All bats were released at the site of capture after recovery from anaesthesia. Serum neutralization tests were conducted on all serum samples at the Australian Animal Health Laboratory, Geelong, Australia, under Biosafety Level 4 conditions. Details on anaesthesia and testing protocols are given in the electronic supplementary material. P.D. conceived and directed the study; P.D. and J.R.C.P. cowrote the paper with assistance from J.H.E.; J.R.C.P. designed and conducted the modelling and statistical analyses with J.D. and A.P.D., and conducted all fieldwork related to livestock; J.H.E. and S.A.R. led the bat survey work, and S.A.R. conducted some of the sample testing. M.B. collected field data on pig production during the NiV outbreak investigation. A.A.J. facilitated data acquisition and all field activities in Malaysia. A.D.H. directed sample testing. H.E.F. assisted in study design, helped direct field activities and provided veterinary input. All authors were involved in the design of the study and the interpretation of the results and commented on the manuscript. All other

Do expert assessments converge? An exploratory case study of evaluating and managing a blood supply risk

BACKGROUND: Examining professional assessments of a blood product recall/withdrawal and its implications for risk and public health, the paper introduces ideas about perceptions of minimal risk and its management. It also describes the context of publicly funded blood transfusion in Canada and the withdrawal event that is the basis of this study. METHODS: Interviews with 45 experts from administration, medicine, blood supply, laboratory services and risk assessment took place using a multi-level sampling framework in the aftermath of the recall. These experts either directly dealt with the withdrawal or were involved in the management of the blood supply at the national level. Data from these interviews were coded in NVivo for analysis and interpretation. Analytically, data were interpreted to derive typifications to relate interview responses to risk management heuristics. RESULTS: While all those interviewed agreed on the importance of patient safety, differences in the ways in which the risk was contextualized and explicated were discerned. Risk was seen in terms of patient safety, liability or precaution. These different risk logics are illustrated by selected quotations. CONCLUSIONS: Expert assessments did not fully converge and it is possible that these different risk logics and discourses may affect the risk management process more generally, although not necessarily in a negative way. Patient safety is not to be compromised but management of blood risk in publicly funded systems may vary. We suggest ways of managing blood risk using formal and safety case approaches.

Blood is a special product, being integral to life and inside our bodies. As Chan [1] notes (following Titmuss [2] ), there is much symbolism associated with blood. As a medical solution to a health problem it depends on the gift of others. Its gift or donation is altruistic and seen as sharing what is essential to life. In Christian societies, blood is seen as having the ability to wash away sins [3] . It represents purity, intensified by its gift to others. Blood donation is a very visible connection to others, even in societies where payments are made to donors [4] . These payments remain controversial and may lead to compromises in the safety of blood products. Blood donation is indeed a two-edged sword: a lifesaver or a transmitter of disease. The struggle over successful compatibility of blood types and to combat transmissible diseases has been a long but largely successful story [5] . The hazards in receiving transfused blood were heightened by HIV/AIDS, which as Chan [1] argues, led to the blood risk being intensified by this stigmatising illness. But since the mid-1980s, screening tests for sexually transmitted infections, including HIV and, later, other transmissible diseases have been implemented in many countries worldwide. In fact, scientific risk assessments, the vigilance of existing or new federal agencies, enhanced public information and disclosure have greatly improved both the calculated and perceived risk of 'bad blood'. As we shall see, Canada is no exception but it is a jurisdiction where there is particular salience. Blood transfusions are overwhelmingly safe, with about 0.5 to 3 percent of all transfusions resulting in adverse consequences [6] . A proportion of these adverse events result from error in preparation and administration. Other adverse events are classified as infectious (e. g. HIV/AIDS, hepatitis, human T-Cell lymphotropic virus (HTLV), West Nile virus) or non-infectious (these are commoner and include acute haemolytic reaction, transfusion associated acute lung injury, allergic reactions, graft-host disease). While severe non-infectious complications are rare, case fatality is high. Infectious disease risks associated with blood transfusion in Canada are currently estimated at 1 per 7.8 million units transfused for HIV, 1 per 2.3 million for hepatitis C, and 1 per 153,000 for hepatitis B [7] . For manufactured plasma-derived products, the risk is estimated at less than 1 in 10 million or theoretical [8] . Bacterial infection from contaminated blood products is also possible but routine bacterial screening of high risk products (platelets) and changes in blood collection techniques have reduced this risk significantly. Methods to inactivate pathogens in blood products (those currently known and future threats), have been developed for plasma and platelets and methods applicable to red cells are under development. Canadian blood suppliers are currently exploring the feasibility of implementing these new technologies once they are available [9] . So despite public concern (see Lee [10] ), experts would surely see risk as minimal and theoretical. And they do. But this does not prevent every 'unsafe' blood incident from being rigorously pursued. But this assumes that all experts may have the same goals or different pathways to similar goals which may shape the treatment of the hazard differently. In part they do -the safety of the product and public health. But do different groups of experts have other goals or interpretations which may shape the treatment of hazard in specific ways? This is the broad question explored in this paper, the major contribution of which is to examine the divergence of expert opinion even where science is consistent and the goal universal. To answer our question, a constructivist approach is adopted. Assessing and managing all kinds of risk is a challenging task but this is especially the case in those risks pertaining to human health which are largely dealt with in the public domain. Because these concerns do invoke public discourse, much attention has been paid to the differences, if any, between expert and public perceptions and assessments [11] . For twenty years or more, constructivist analyses have brought to the fore the importance of prior social, cultural, institutional and political factors that shape and are embedded in both lay and expert risk assessments (see [12] ). Lay people and experts use similar cognitive processes with both falling prey to errors from anchoring, overconfidence and the gambler's fallacy (see [13] ) Attention has also turned to the perceptions and assessments of experts themselves. Van Zwanenberg and Millstone [12] comment that the construction of scientific claims can inform risk assessments. Furthermore, a "coherent realm of expertise" may not exist, as Haggerty [14:201] points out with respect to crime risk. He adds that expert opinion about crime risk is incredibly fractured with there being differences in professional opinion with respect to dimensions of high-profile risks as well as the nature and efficiency of crime prevention. Sjoberg [15] also points out that there is likely to be a whole range of expertise and this range is not well-articulated in many risk perception studies. If, therefore, there is this range of expertise, why should we expect expert opinions to converge, although overconfidence in technical and medical assessments may exist (but see [16] )? In fact, we shall note certain biases in expert opinion. Shanteau [17] questions the convergence of expert opinion which is based on the widespread use of statistics and economics in assessments and on the widespread search for generalizability in how people think about risks. Shanteau argues then that the bases of convergence are in themselves flawed (see also [18] [19] [20] ). Of course, it is recognised that different risk logics or discourses -i.e. different conceptions of an activity or event as risk -exist, but these have seldom been applied to expert opinion (see [21, 22] ). Silva et al. [23] show, in an experimental setting, that despite differences in scientific background and political beliefs, scientists tend towards a precautionary stance over the setting of safety standards. Yet scientists are not the only type of expert involved in such matters. McMahan et al., [24] point to the differences in how scientists and risk assessors regard electric and magnetic fields. Furthermore, Chalmers et al. [25] show how the opinions of dentists and nursing directors vary over dental care in nursing homes. Chociolko [26] provides an example of expert disagreement often found in sometimes adversarial settings in which experts 'take sides' representing, say, industry or community. How might we assess expert opinion and convergence on risky matters? Shanteau [17] points to the importance of the level of decision, made by experts using a medical analogy of diagnosis (what is it?), prognosis (what is the likely outcome?) and treatment (what to do about it?), with a different logic applied on different levels. Furthermore, experts and professionals possess similar cognitive properties as non-experts. "Experts are not immune to the cognitive illusions that affect other people" [27] . And it is increasingly noted that cognition and emotion are intertwined in decision-making (see [28] ). As Cross [29:28] comments: "when an attitude or an activity is of considerable importance to a person, the individual is loathe to believe that it is hazardous." Furthermore, biases or heuristics found in studies of perceived risk may be found among experts and affect their assessments and opinions. In this paper, we address some of these issues about expert opinions, utilising the ideas of non-convergence (especially at different levels of an assessment), and of the heuristics made by different professional groups (commitment to a particular practice, reliance on rational models and thinking), to understand different concepts of risk relating to the blood supply arising from a case when supply of 'safe blood' was potentially disrupted in one Canadian city. So all may want 'safe blood' but do the management and decision-making styles and discourses of all actors converge or vary? The Context and the Case 'Safe blood' has been a major policy issue in Canada. This is also a sensitive issue, which likely arises from the 'tainted blood tragedy' of the 1970s and 1980s. At that time, the blood supply was organised and managed by the Canadian Red Cross Society (CRCS) as this body had been responsible for blood supply during the Second World War. CRCS is a not-for-profit, humanitarian society, with many diverse activities extending beyond the responsibility for the blood supply which are largely carried out by volunteers. A chronic lack of funding for the CRCS held back technical developments to collect and supply blood in Canada. Government funding only began to increase in the mid-1970s as procedures for blood collection became modernised. But its principles as part of the international Red Cross (non-discrimination against individuals and independence from government) meant its screening and management procedures became fatally flawed in the Canadian context, with often strained relations between volunteer oversight and professional activities. These limitations set the stage for tragedy with the discovery of blood-borne diseases as recruitment of donors was still in volunteer hands and carried out with self-answered questionnaires. The recipe for disaster was set, and the situation was worsened by the failure to track those who had received tainted blood and a failure to apologise or compensate affected individuals on the parts of the provincial and federal governments (see [30, 31] ). As Picard [30] has noted, 'tainted blood' was arguably the largest public health catastrophe in Canada. "About 1,000 individuals who received blood transfusions in Canada between the late 1970s and 1980s were infected with HIV and another 30,000 were infected with hepatitis C" [32] . Eventually, this led to the establishment of a public inquiry in 1993 with the final report of the Krever Commission being delivered in 1997 [31] . Krever recommended the creation of a new blood system and the Commission's recommendations led to the creation of Canadian Blood Services to operate the system in all provinces, except Quebec which formed its own agency, Héma-Québec. With the creation of these agencies, the Canadian blood system joined the U.S. and West European countries in having an expert-based and scientifically-based system and CRCS went on to different humanitarian tasks. For the blood system and perhaps other aspects of the Canadian healthcare system, Krever highlighted two key elements -the institution of precautionary measures and the creation of a governance system emphasising safety (see [32] ). Risk assessment became based on scientific tests and evidence and the management system highly coupled and structured to protect human health. The Canadian blood system is currently seen as a high reliability organisation, emphasising safety and responsiveness to problems [33] . In fact, these elements were introduced so the Canadian blood system could respond to threats in a timely and effective way. This was the case for threats from infectious agents with product recall or donor deferral being rapidly instituted for variant Creutzfeldt-Jacob disease (See [33, 34] ), West Nile virus (see [35] ) and SARS. It is also the case for 'organisational threats' in which practice errors over labelling, documentation, or other deviations from procedures are dealt with through product withdrawal. For such threats, it may be said that the Krever report is one dimension in the increasing use of quality assurance for among other things patient safety in health care. But we know from other settings that challenges occur and 'accidents' (often resulting from operator error or technical malfunction) are normal in complex systems [36] . It seems therefore worth discovering if in this highly coupled, apparently well-managed system, all experts agree about the nature of arising hazards. Recalls involving small numbers of blood products occur frequently (daily), for example, when donors provide post-donation information that affects their eligibility as a blood donor. Whereas withdrawals, which occur less frequently, are often due to operational deviations where the risk is minimal or unknown and may involve a large number of blood products. Withdrawal involving a large number of blood products can jeopardize the availability of an adequate blood supply. When there is a real or perceived threat to the safety of the blood supply, withdrawals occur without reference to cost or potential short-term shortages. A major withdrawal of blood products in Ontario in 2005 involved over 3500 blood product components. The reason for this withdrawal was clerical in nature associated with records of donation which are completed at the time each donor gives blood. The risk to the blood supply was likely minimal or nonexistent. Individuals in the appropriate organizations joined forces to deal with the matter. But it brought to light many of the challenges faced by both hospitals and the blood suppliers around dealing with error and emphasized gaps in the system related to timeliness; communication; risk perception; and recipient notification. It provides an example of what we call organizational 'threat' or, more positively, a learning opportunity. We suggest that this approach to identifying error as a potential for hazard and risk is useful in that it focuses attention on possible differences in organizational responses and public notification. So given the context, it would appear that the case could have been dealt with procedurally with all parties -regulators, blood product supplier, hospitals and treatment facilities and agents -being on the same page. The risk from the blood product should have been seen in a universal way by all parties involved in this particular case. But was it? If there are different levels of decisionmaking and risk assumptions do expert assessments necessarily converge? The answer, to anticipate, is yes and no. This paper is an outgrowth of a study undertaken to determine the current procedures and protocols for handling the recall/withdrawal of blood products in Ontario and beyond [37, 38] . In this we adopt a quasifoundational position, with the experiences and views of respondents being seen in the answers to our questions (see Additional File 1: Appendix 1) from which exemplars in the form of quotations are drawn (see [39] ). We undertook a number of one-on-one interviews with key stakeholders in the blood supply and management arena, both within and beyond Ontario. Interviews were conducted with individuals from Canadian Blood Services, Héma-Québec, Ministry of Health and Long Term Care (Ontario), relevant staff at hospitals and hospital laboratories, as well as blood product recipients. These recipients were not professional or credentialized experts but it was felt that their experience of the blood transfusion system might add a different perspective to our investigation. Within hospitals, we interviewed physicians (both transfusion medicine physicians and physicians who do not work in the transfusion service but who use blood products), hospital administrators, CEOs, risk managers and public relations personnel. Within hospital laboratories, we interviewed managers, technologists and transfusion safety officers. In other words, we interviewed in total 45 professionals involved in blood supply -regulators, suppliers, treatment individuals and agencies, administrators, laboratory personnel and transfusion recipients. The project's steering committee (composed of individuals from the blood supplier, hospitals and laboratories) helped to identify individuals from the blood supply and management arena who had experience of blood product recalls and withdrawals and who could therefore be anticipated to be able to provide rich data. This purposeful sampling was augmented by asking those individuals interviewed to suggest others whom they thought it would be helpful to interview. We employed a multilevel sampling framework in that we wished to compare the perceptions of those with potentially different stances on blood risk management (see [40] ). We recognise that the sample size in our study limits the generalisability of our findings and interpretations, and it may indeed be exploratory with our practice suggestions being essentially that-suggestive. Table 1 shows the number of participants interviewed within each category. As well as interviews, copies of written relevant rules and procedures pertaining to recalls/withdrawals were obtained. And we may askdoes opinion converge on the type and nature of the risk? We also planned to conduct interviews with individuals from the Ontario Hospital Association and from Health Canada (the regulator of the Canadian blood system). It became clear while talking to individuals from the Ontario Hospital Association that the association does not currently play a role in blood product recalls/withdrawals within the province. Health Canada, despite being asked on several occasions, declined to participate in the project, having been advised by their legal department not to do so in order to avoid the possibility of liability issues. Recipients of blood products 2 Note: Numbers in the text after job title refer to individuals in that category interviewed. In order to obtain as comprehensive an understanding as possible of the recall/withdrawal process across the province of Ontario, we made sure that the different types of hospitals were represented (large urban teaching hospitals, smaller urban hospitals and rural hospitals). We also interviewed individuals from Héma-Québec, to get a sense of how the process compares between blood suppliers and key informants from other provinces (Alberta, British Columbia, Nova Scotia, and Saskatchewan). The protocol was approved by the Research Ethics Board of McMaster University. The interview guide included questions on understanding the terminology involved in recall/withdrawal situations; individuals' experiences of these situations and the actions taken by different individuals and or stakeholders at different stages in the process; existing policies and procedures; questions related to disclosure of information and notification of recipients, including who should be involved in this process and how it should be done. Respondents were also asked for any suggestions they might have on improving the process (see Additional File 1: Appendix 1). Interviews were conducted over a four month period from May to August 2006, either in person or by telephone. The interviews were conducted by two of the research team members (EA and BMcC). To maximise consistency of questioning, a semi-structured interview guide was followed in each of the interviews. The questions were open-ended and interviews lasted between 45 minutes to 1 hour. All interviews were audio-taped and then fully transcribed. The transcriptions were checked for accuracy against the taperecordings and then imported into NVivo 7, a qualitative data management software program commonly used in qualitative research [41] . A team analysis approach was employed whereby as transcripts became available, they were read independently by several members of the research team (EA, BMcC, and JE). The team then met at regular intervals to discuss the content of interviews and to identify the themes and issues emerging from the data. A schematic for coding the data (identifying discrete passages of text that contain the same idea) was developed by EA and BMcC and then this coding scheme was applied to a new batch of interviews by EA, BMcC and JE. Interrater agreement was then calculated and was found to be very high (100% for major codes and 94.3% for minor codes) thus indicating that the coding scheme was working well. This schematic was then used to code the entire data set by BMcC, so that all interviews were coded using the same coding themes. This schematic acted as a taxonomy, classifying and organizing the complexity of the 45 responses and there is a close parallel between the taxonomy (reported in [38] ) and the interview guide. Further, deeper coding of the data generated other propositions, which may be seen as second order constructs (see [42] [43] [44] ). The analytical development of these constructs allowed relationships to be identified between the codes in the taxonomy (see [45] ). These also emerged because of our interest in blood risk management. Thus the issues of notification, response to minimal risk, responsibility and legal requirement are transformed into those appearing in the results section, guided by themes and theories of risk management and expert judgement outlined above, i.e. the heuristics and logics used to deal with uncertainty and to manage risk for protective, liability and precautionary reasons. In presenting the findings we have selected quotes that illustrate these ideas, allowing as many respondents as possible to speak. We suggest, therefore, that our paper makes a methodological contribution by using the typifications of social phenomenology to transform respondents' concerns into specific risk discourses. As a foreshadowing to outline different conceptions of how blood product risk should be managed, we note confusion among respondents over the meaning of the terms 'recall' and 'withdrawal'. There is meant to be clarity in: "With respect to a health product, other than a medical device, means a responsible party's removal from further distribution or use, or correction, of a distributed product that presents a risk to the health of customers or violates legislation administered by Health Products and Food Branch Inspectorate (HPFBI)" [46] . Withdrawal "The removal from further distribution or use, or correction of a distributed product where there is no health and safety risk and no contravention of the legislation administered by the HPFBI. It is not considered to be a recall" [46] . Yet only 14/45 (30%) of individuals interviewed indicated that they knew the difference between the two terms. The difference between the terms is confusing for both hospital and blood supplier personnel: "I think they are confusing. The only reason, I'll be honest; the only reason that I'm familiar with it is because of the incident that we went through ..." (Laboratory Manager-01) "Well ... it's very confusing. We are not exactly sure when they recall something or withdraw something. It's, you never know why they are asking you to return something. And they don't give you any extra information, and it is somewhat frustrating, because you know why are they doing this?" (Lab Technologist-06) "Um... gosh... there probably is a very important difference and I ... I would be lying to you if I said with confidence that I could tell you the difference" (Physician using blood products-01) The confusion is mainly associated with diagnosis (what is it?) and does not appear to influence the action (what is to be done?) taken to deal with the recall or withdrawal notification at the hospital level. Regardless of whether a recall or withdrawal is issued, the initial action taken by the hospital Transfusion Service is the same: implicated products are removed from useable inventory. So would uncertainty be removed if one term was applied? "I don't think they should be handled differently. You know if for whatever reason a product is being taken off the market, it would, it's unfortunate that you have two terms where again they have different connotations. I think one term should suffice for all and... they should be handled in the same way." (Physician Blood Bank-03) "Well I guess the fact that we have so much trouble remembering which is which... could be problematic. I think for the regulator they do require. I mean it's important to have two different terms... with two different definitions because they, they are two different matters. In practice that we get the two terms mixed up, I'm not sure it really matters..." (Blood Supplier-07) While there were differences in labelling what was happening, there were none in terms of prognosis (likely outcome), seen universally as the removal of unsafe product. In other words, expert opinion converges with respect to the goal -safe blood -but on how and why this might be done there is divergence, thus revealing differences in management perspectives. We identify three approaches to managing the risk from this organisational threat: risk as hazard, risk as liability and risk and precaution. Risk as hazard emphasises the potential adverse consequences to patient well-being. Such an approach demands immediate removal and the full disclosure of what is happening to patients. Furthermore, it suggests that those notifying patients should be as close to the patients as possible, usually the treating or transfusing physician. This risk is almost a given and most comments refer to the importance of physician notification to allay any or all fears about hazard, the physician being seen as trusted, knowledgeable and close to the patient. There is agreement on who is central in this immediate task of ensuring blood safety, the blood supplier, but different stakeholders hold the risk as hazard view for different reasons. The blood supplier itself sees a distinctive chain of responsibility and action if blood may be unsafe -from themselves to the physician to the patient. "... you know we are in this era of informing the patient, but I think to some extent the pendulum can go too far ... and that we need to be careful about not giving patients information that's no of value... I think we can overdo some of the informing of patients, it's almost like we're passing the buck and not kind of letting the responsibility stop somewhere along the way with a physician..." (Blood Supplier-07) This discussion on the need to notify seems supported by all stakeholders. Hospitals see that any notification from the blood supplier highlights a concern and the need for action. "Now, I would anticipate that if the blood... supplier... was concerned enough to notify us as an organization to recall a product, then the degree of risk is always such that it would be important for us to notify the patient. You know what I mean. Like, the risk assessment has already taken place at the blood supplier" (Hospital-Risk Management-02) "I guess from my perspective there is either risk or there is no risk, and if there's no risk they're not notifying us. If they're notifying us it's because there is risk" (Hospital-PR-01) Physicians tend to agree but see themselves and are seen as those best positioned to make patients aware of potential problems as 'they know' their patients best. "In my opinion, it's the role of the clinician who's caring for the patient to manage those things" (Physician using blood products-05) "It's useful to have a well-informed recommendation from the Blood Supplier but the hospital always has to use its discretion and the difference there is that we know our patient population" (Physician using blood products-07) "I think ultimately the physician always has that, you know, discretion. And that's clear in case law" (Blood Supplier-04). Others feel that the physician may not have all the skills necessary to manage risk as hazard, especially if their style is based on a confident, medical approach to the issue. "So I would be perfectly supportive, of a hospital system or provincial system or even a nation-wide system that developed advisory guidelines. But I don't think it's a situation where if it is ... that a physician doesn't want to notify her patients that, that somehow some other agency needs to get involved and compel that notification" (Physician-Blood Bank-03) "I don't know whether there would be a place for a third party to come in, where a third party would contact the recipient and say 'I know that your physician spoke with you, or that you received a letter, this is just a reminder that you might need to go for further testing'" (Laboratory Technician-04) "Well I don't know if everybody has them but a patient representative would be one option. I think somebody with those sets of skills. Like not just the technical skills but the people with counselling skills and discussion. I mean they'd have to have some kind of knowledge about blood and risk" (Provincial Health Ministry-02) Risk as hazard was the best articulated of the discourses on how to manage. Yet risk as liability, where the potential consequences for system integrity of specific practices were central, received greater expression from respondents. Patient well-being is of course still vital but now disclosure to patients of events that might affect their wellbeing protects the blood supply institution as well. Such management is seen particularly at the supplier and hospital levels. The identification of the risk and its disclosure were closely related. But under many comments lies an often implicitly stated concern over liability, i.e. what is our liability if we do not act (disclose) and something happens? In other words, their responses are anchored around current operational characteristics of the healthcare system and their legally demarcated roles within that system. Liability is often framed in terms of outside perceptions, specifically and not unusually, on how public perceptions of expertise and expert response may be framed. In this discourse, the role of the blood supplier is central but responses to their notices may vary depending on how others see the issue and their liability. The blood supplier must always give an opinion. "We [Blood Supplier] would always give our opinion as to follow up." (Blood Supplier-01) The importance of local circumstances and their possible impact on liability are recognised by the blood supplier. It must balance its obligations with that of the independence (and discretionary action) of institutions such as hospitals and doctors. "What we do in our centre is... we take a look at the reason for the recall, we may provide that information. The hospital doesn't do much with it, and they know they'll get a supplementary letter from me if I think recipient notification is required so that decision whether recipient notification is required is made here. I think that ... works better in our environment because I have more experience in this than the regional lab tech ... and we don't have experienced blood bank directors in most of the blood banks." (Blood Supplier-02) Yet circumstances at the local level may affect response to this opinion. "But the Blood Supplier has their own ideas about what, in what situations do recipients need to be notified so they advise us. In our opinion, you do or do not have to notify the recipient if this product has been transfused already and we sometimes do what the Blood Supplier asks, although we're a little more aggressive about notifying patients than what the Blood Supplier does." (Physician using blood products-01). "I think you know, some hospitals are in a position where they absolutely must rely on the expertise of the Blood Supplier. I mean, they, you know, primary care hospitals that I assume are somewhat more comfortable just saying, 'Look, just tell us what you want to do. Tell us what you want to say and we're not gonna do any independent analysis. Just, you know, provide us with what your recommendation is.' Other hospitals are saying, 'okay Blood Supplier, thanks for the information. We're gonna consider this independently. We have the expertise to do so, you know. The only thing we want you to do is provide as much information as you can on the risk and we're all consider it at our Transfusion Committee and we'll all decide ultimately what we think our physicians should do in terms of patient notification.' ... that kind of thing. You know, so there's quite a difference. So that is a part that we're struggling with a little bit in terms of how do we fulfill our obligations? We never want to be, we never want to fetter anyone's discretion in terms of notification. That's for sure. Though part of us is saying, you know, part of the time we think, okay, we can provide all of the, all of the information, all of the risk information as clearly as we can, and that's it and then the hospital can sort of make their decisions. But there's another component then because we don't want to be in a position of... we don't want to abdicate our responsibility and thrust the decision making on hospitals, you know, so there's sort of a tension there between those two" (Blood Supplier-04). The hospital is perhaps in a position of having to respond to local pressures before the full facts are known because they are local institutions and have explicit liability concerns. "I believe in full disclosure even if all you can say is here's what we think happened, here's why it happened, here's what we know and don't know, and here's a mechanism for either monitoring or what have you down the road. So in the absence of that, in the event that there's some new novel research finding in the not so distant future, I've never been told about this, I didn't know I had exposure, some marvellous new technology or technique comes along that would perhaps allow me to be more definitive. Well I don't even know about it to pursue that or include it in my own medical history" (Hospital Adminstrator-02) While some groups which are closest to the patient that is being transfused with blood see most recalls as leading to minimal risk, perhaps displaying an overconfidence in the role of scientific assessment, others are concerned about outside perceptions and what these might do to the situation. " ... we couldn't identify what the risks were and the risks were minimal, therefore we should not disclose, and [the Chief of Governance] was adamant that we needed to ... I was really surprised that the recommendation by the Transfusion Committee was totally disregarded" (Transfusion Safety Officer-01) Waiting for an assessment can lead to much time thinking through the issue. This may result in an affect response, especially with respect to the media. This may be exacerbated by public attention being roused before a system-based announcement can be made. "And sometimes we've got no, we don't have an assessment yet... so we're sitting ... waiting, you know and there's a time delay there so... my biggest fear is in the meantime it hits the media, now we've got an issue" (Laboratory Manager-03) "But the problem is, is that you have media that's watching, and so then, then there's a different spin put on this when it hits the newspaper. You know... another bloody scandal. So are you more concerned about your public relations or are you more concerned about the effect you're going to have on patient care?" (Laboratory Manager-01) Much of the liability discourse takes into account the perceptions and roles of other stakeholders, both inside and outside the blood supply system. The third discourse -risk as precaution -does that too but also seems to treat problems on a case-by-case basis as uncertainty cannot be fully removed or explained. It is a view of risk often held and articulated by system administrators. A precautionary approach has been based on a changed perspective toward patients and institutional/ professional partners. And precaution -taking care -is necessary as interests and ways to achieve them may be dissimilar. The blood supplier sees precaution as necessary because of these dissimilarities. "I don't know how much a hospital's disclosure policy would vary from one hospital to another. I mean one would hope that there's some uniformity in that or else disclosure of things is going to be a problem for very much more than recalls and withdrawals because I think there's all kind of things in a hospital that one might decided you disclose or you don't disclose" (Blood Supplier-07) "I'm not looking at it in the capacity of disclosure and our disclosure policy and for us it's not as simple as just telling people something went wrong. We have to ... weigh out what the risk of telling someone versus the benefit of telling someone ... we very much are very strong proponents of disclosure and do it unless the risk of advising is significantly worse than not we certainly learn towards advising patients" (Hospital Administrator-03) Furthermore, the blood supplier argues for precaution because of the need to respect the different needs and sensibilities of different types of patient. "Because right now, the decision whether to inform a patient is based upon the doctors and the doctors in hospitals alone. The patient has no input whatsoever.... now the CJD thing was a perfect example of some people probably didn't want to know all the recalls, because what good could this have done. But some people did. So that there has to be a choice made by the patients... and my view and most patients' view on that is that they have no business making that decision for patients. Now, it's just the whole attitude of the health system, it's not anyone's particular fault, but it's just the attitude of you know, it's your responsibility and we'll make the decision. I think patients feel you know a little ticked with that" (Transfusion Recipient-02) Often, agencies lack information about these patients: the recipients of the transfused products. "We only have part of the story when we do these recalls and withdrawals which is the information from our end about the donor or about the component but we don't know anything about the recipient at the other end. And the importance of the information and about what should be done has to be determined in the context of the recipient" (Blood Supplier-03) "I know our medical staff has a problem with the word recommendation because, I think it centres around the fact that they are dealing with facts about the unit and have no facts about the patient. Certainly, if you go to the level of individual patients, you can, I think justify different actions because of the different situations of patients" (Blood Supplier-01) Precaution on the part of local system administrators is weighed not only with respect to hazard but also institutional and patient autonomy. "For us it's not as simple as just telling people that something went wrong. We have to, we like to, weigh out what the risk of telling someone versus the benefit of telling someone ... so I think there needs to be a way to bring hospitals together and recognising the hospitals are independent and are free to make their own choices to the extent that that can be coordinated goes a long way because ... we have to have a plan B that while we didn't think disclosure was appropriate, in the event that another hospital chose to disclose, we had to be prepared for how we would respond to that" (Hospital Adminstrator-03) "It's a very paternalistic approach to say 'oh well we know it's bad for you and so you know we want to spare you the pain of having to you now think about things like this so just let us do it.' That has not worked well in the past and we are a sort of newer generation of people where the attitude has changed... So, give the sort of transition in the social more that is out there.... I think trying to take that paternalistic approach that we will hide things from you for your own best interest it just won't sell" (Transfusion Recipient-01) In managing the possible consequences of this error, a quite minor, almost theoretical, risk to the safety of the blood supply, there was uncertainty about what was being managed -a recall or withdrawal. This has now been resolved in part through the authors' report (see [38] ). The term recall is now used. Clarity in identifying responses to a possible hazard is necessary. But the then uncertainty around terminology did not stop the problem from being tackled but it caused pause for thought. We note too in this intensely regulated, safetyfirst environment that conventional risk discourse -as hazard -and management as its removal or mitigation did not loom large. It is a given in this tightly knit and collegial community that patient safety is paramount. There appears to be significant levels of trust between the different groups of professionals, something not always found in hazard management settings (see [47] ). This trust and respect provide an excellent basis for existing and enhanced communication about blood hazards in these groups. Communication between parties with respect to the issue, what it means, and what responses are possible is key. It remains important to remember that there may be differences in approach to, say, discourse, or treatment and some adjustments may be required. And while the discourses are ones of engagement in this case, that does not necessarily mean that there will not be adversarial interactions over preferred management strategies and rationales, especially in the importance given to patient notification. (See [48] for a discussion of expertise and collaboration). This need for disclosure may in fact lead to a further risk management challenge. In our case, there was discussion among risk managers about risk amplification through disclosure. Research [49] has certainly shown the importance of the physician communicating risk issues to the public. Risk as liability views system integrity and maintenance as an important goal. Management entails not only the use of knowledge about patient safety and system practice but also a cognitive and emotional commitment to the aims and goals of the organisation. The supplier agency has created that commitment and loyalty within a changing Canadian health care system (see [37] ). Yet uncertainty remains and this may be due to the centrality of precautionary principles in the blood supply system. Thus, it is not surprising that risk and precaution are seen simultaneously as two dimensions of hazard management. Good record keeping and monitoring of transfused blood (and its recipients) enables precaution to be central in risk management. Kaplan et al [50] advocate a medical event reporting system for transfusion and we concur with their suggestions. Such a system (Transfusion Error Surveillance System -TESS) is currently being developed and piloted in Canada [9] . Furthermore, the difference discourses in expert groups within the same decision-making system (but likely at different levels of decision-making) may be beneficial, if recognised as such. 'Hazard' is a safety-first approach protective of health; 'liability' ensures that the challenges of risk amplification and perceived injury may be dealt with if an unknown risk is disclosed and 'precaution' ensures due diligence and quality assurance before action is taken. With respect to the three risk discourses, some are more likely to be articulated by some professional groups -hazard by physicians, liability by administrators (hospital and blood supplier), precaution by administrators (especially the blood supplier). Those groups are largely responsible for managing the risk from these respective vantage points. There is largely a coherence of expertise within those realms, although it is in part challenged by laboratory technicians (hazard and liability) and transfusion recipients (liability and precaution). A vital strategy in managing any risk or consequences of error is a coherent response. This coherence may break down as different dimensions of risk management are required (patient safety, system maintenance) or different professional and lay groups engaged. In fact, it is possible for different discourses to be used by one or more different groups. In all discourses, it is possible for different heuristics to guide management response, overconfidence with respect to medical expertise and the scientific assessment of risk (physicians and blood supplier), anchoring to the legal and formal positions of institutions (hospitals), and affect with respect to feelings about uncoordinated public announcement of a risk and perceived media response (laboratory technicians and transfusion staff). The blood supplier uses all discourses because of its variable role in supplying safe products. In adopting a quasi-foundational approach, there are ways to ensure rigour and the trustworthiness of interpretations. The social phenomenology of Schutz, used to derive themes and as a basis for theory development from respondent perceptions, requires meeting three postulates, namely logical consistency, subjective interpretation and adequacy. For the first, we have highlighted how the research problem and methods were derived from a real world problem on which was based the questionnaire, sampling strategy and the need to interview; for the second, by using respondents' views to develop interpretation (different than most studies as the first order or naturalistic constructs are expert ones); and for the third, by linking second order constructs to activities and phenomena in blood risk management. Furthermore, trustworthiness can be demonstrated by credibility and confirmability (see [51, 52] for practice-based examples). We have provided a trail from problem identification, questionnaire development, coding taxonomies to themes and theoretical development. We have also provided the rationale of respondent selection and sampling design as well as an outline of procedures followed to arrive at constructs or themes. We have derived through these themes risk discourses and logics relevant to blood risk management. Before finalization, the study results and recommendations were presented at a consensus conference of study participants and stakeholders, to ensure validity of data interpretation. All suggestions made were incorporated into the final results. Although we recognize that all researchers will not necessarily agree to the ways we have sought validity for these findings (see [53] ), first author conversations with risk managers point to the utility of the interpretations. Thus this qualitative investigation has contributed a nuanced risk characterization. In fact, all discourses are necessary to manage this low-level risk. This coherence and differentiation of expertise around managing blood risk has practical consequences. As Hunt [54] notes, governance of risk is characterised by risk avoidance rather than risk management and is in turn dominated by a preoccupation with safety. The further corollary is the expanding panoply of regulation and guidelines. This may be noted in blood supply. And when 'something happens'-a threat to safety-the virtuous cycle of risk, regulation, and prescription is interrupted. The relentless pressure for the systemisation and integration of risk management practices continues as a watchword for corporate social responsibility (see [55] ). This virtuous cycle is reinforced by the use of a precautionary logic. As Haggerty [14] , notes, precaution emphasises the worst eventualities. It is not so much about risk. It "invites one to anticipate what one does not yet know, to take into account data, hypothesis and simple suspicions" [56:288] . With a product as vital and special as transfused blood, precaution is necessary. But societally it may feed anxieties and increase risk aversion. Practically, we point to consideration being given to enterprise risk management (ERM), recently developed as ISO 31000 (see [57] ) and accepted by the Canadian Standards Association in 2010.. ERM considers any risks or uncertainties affecting objectives, requires a flexible organization to tailor risk management, formalising monitoring review and consultation, and demands accountability from all those dealing with the risk. All senior managers must be committed to the process which must be used by all decision-makers in a flexible organizational structure. In its stages, it considers risk in careful ways. In setting the context and identifying risk, it suggests consideration of risk appetite and triggers. In analyzing and treating risk, ERM points towards acceptance, control and mitigation. It suggests ensuring that residual risk and its potential impacts are not ignored as it is not possible to remove all uncertainty. ISO 31000 may be complemented by using a safety-case approach which requires the incorporation of all evidence to ensure the system is safe to operate (see [58] , potentially modified to permit inputs from all appropriate stakeholders to organize heterogeneous information and concerns to ensure the safety and dependability of a larger network (i.e. the blood system). Elements of the blood management system have been drawn to ISO 31000. Furthermore, processes such as recording, inter-professional communication, notification and disclosure have been applied in dealing with blood risk. Risk triggers and residual risk point to the relevance of hazard and liability discourses, especially as in other domains, use of the precautionary principle has been shown to trigger concerns and lower trust in governance structures (see [59] ). So for the institution of precaution we must be clear on what we are managing, why and in what ways. For the Canadian blood system, a new procedure of providing reasons for recall will assist hospitals and other donation agents in managing perceived risks. Yet the role of expert biases and domain interests are likely to continue to exist, and must be understood and incorporated to ensure blood safety and continued public trust and to provide timely responsiveness in such a high reliability system. We suggest the framework of ISO 31000 emphasising context, risk identification and assessment, risk treatment, communications and consultation is useful, along with a safety-case approach. Communication about ways to protect public safety is always necessary and this must include clarity on definitions, responsibilities, and public perceptions and what the consequences of even minor errors mean in a complex system. Furthermore, error as a risk state needs careful theorizing and application in systematic risk management approaches. Additional file 1: Appendix 1: Interview guide for understanding the management of blood products in Ontario. Appendix provides the questionnaire used to explore risk management of blood products under conditions of uncertainty.

Noninvasive positive pressure ventilation for acute respiratory failure in children: a concise review

Noninvasive positive pressure ventilation (NPPV) refers to the delivery of mechanical respiratory support without the use of endotracheal intubation (ETI). The present review focused on the effectiveness of NPPV in children > 1 month of age with acute respiratory failure (ARF) due to different conditions. ARF is the most common cause of cardiac arrest in children. Therefore, prompt recognition and treatment of pediatric patients with pending respiratory failure can be lifesaving. Mechanical respiratory support is a critical intervention in many cases of ARF. In recent years, NPPV has been proposed as a valuable alternative to invasive mechanical ventilation (IMV) in this acute setting. Recent physiological studies have demonstrated beneficial effects of NPPV in children with ARF. Several pediatric clinical studies, the majority of which were noncontrolled or case series and of small size, have suggested the effectiveness of NPPV in the treatment of ARF due to acute airway (upper or lower) obstruction or certain primary parenchymal lung disease, and in specific circumstances, such as postoperative or postextubation ARF, immunocompromised patients with ARF, or as a means to facilitate extubation. NPPV was well tolerated with rare major complications and was associated with improved gas exchange, decreased work of breathing, and ETI avoidance in 22-100% of patients. High FiO(2 )needs or high PaCO(2 )level on admission or within the first hours after starting NPPV appeared to be the best independent predictive factors for the NPPV failure in children with ARF. However, many important issues, such as the identification of the patient, the right time for NPPV application, and the appropriate setting, are still lacking. Further randomized, controlled trials that address these issues in children with ARF are recommended.

Breathing difficulties are common symptoms in children and common reason for visits to the emergency department [1] . In United Kingdom, respiratory illnesses (both acute and chronic) accounted for 20% of weekly general practitioner consultations, 15% of hospital admissions, and 8% of deaths in childhood in 2001 [2] . Although the great majority of cases are benign and self-limited, requiring no intervention, some patients will require a higher level of respiratory support. Invasive mechanical ventilation (IMV) is a critical intervention in many cases of acute respiratory failure (ARF), but there are definite risks associated with endotracheal intubation (ETI) [3] . By providing respiratory support without ETI, non-invasive positive pressure ventilation (NPPV) may be, in appropriately selected patients, an extremely valuable alternative to IMV. It is generally much safer than IMV and has been shown to decrease resource utilization and to avoid the myriad of complications associated with ETI, including upper airway trauma, laryngeal swelling, postextubation vocal cord dysfunction, and nosocomial infections [3] . NPPV usually refers to continuous positive airway pressure (CPAP) or bilevel respiratory support, including expiratory positive airway pressure (EPAP) and inspiratory positive airway pressure (IPAP), i.e., biphasic positive airway pressure (BIPAP) and bilevel positive airway pressure (BiPAP), delivered through nasal prongs, facemasks, or helmets. Although there is high-level evidence in the literature to support the use of NPPV for the treatment of ARF due to different causes, such as exacerbation of chronic obstructive pulmonary disease [4] and acute cardiogenic pulmonary edema [5] in adults, there are few reports about its use in this acute setting in children. So far, case series constitute the vast majority of the available knowledge in this age group. However, there is an increasing interest in the use of NPPV as a therapeutic tool for children with respiratory distress that is clear from the increasing number of published studies over time ( Figure 1) ; a research of studies on the use of NPPV in children > 1 month of age, published before December 30, 2010 (database: MEDLINE via PubMed; keywords: noninvasive ventilation, non-invasive ventilation, noninvasive positive pressure ventilation, non-invasive positive pressure ventilation, bipap, continuous positive airway pressure; age limits: children from 1 month to 18 years old) identified 332 relevant articles, of which 48% were published during the past 5 years. This concise review is designed to focus on the effectiveness of NPPV in children > 1 month of age with ARF (excluding patients with neurologic or chronic lung disease). The frequency of ARF is higher in infants and young children than in adults. This difference can be explained by defining anatomic compartments and their developmental differences in pediatric patients that influence susceptibility to ARF [6] . In addition, respiratory failure often precedes cardiopulmonary arrest in children, unlike in adults where primary cardiac disease often is responsible. Therefore, prompt recognition and treatment of pediatric patients with pending respiratory failure can be lifesaving [6] . Respiratory failure is a syndrome in which the respiratory system fails in one or both of its gas exchange functions: oxygenation and carbon dioxide elimination. In general, patients with respiratory failure may be classified into two groups, depending on the component of the respiratory system that is involved: hypoxemic respiratory failure and hypercapnic respiratory failure [7] . Hypoxemic respiratory failure (known as type I) Hypoxemic respiratory failure (type I) can be associated with virtually all acute diseases of the lung, such as status asthmaticus, bronchiolitis, pneumonia, and pulmonary edema, which interfere with the normal function of the lung and airway. The predominant mechanism in type I failure is uneven or mismatched ventilation and perfusion (intrapulmonary shunt) in regional lung units. This is the most common form of respiratory failure, characterized by a PaO 2 < 60 mmHg with a normal or low PaCO 2 . The primary treatment of type I respiratory failure in children is to administer supplemental oxygen at a level sufficient to increase the arterial oxygen saturation (SaO 2 ) to greater than 94%. In situations when a fraction of oxygen in inspired gas (FiO 2 ) of greater than 0.5 is necessary to achieve this goal, this often is referred to as "acute hypoxemic respiratory failure" [7] . In this setting, NPPV may be considered. Hypercapnic respiratory failure (known as type II) Hypercapnic respiratory failure (type II) is a consequence of ventilatory failure and can occur in conditions that affect the respiratory pump, such as depressed 1993 1993-1995 1996-1998 1999-2001 2002-2004 2005-2007 2008-2010 Time years References (n) neural ventilatory drive, acute or chronic upper airway obstruction, neuromuscular weakness, marked obesity, and rib-cage abnormalities. Alveolar hypoventilation is characterized by a PaCO 2 > 50 mmHg [7] . The onset of type II failure may be insidious and may develop when respiratory muscle fatigue complicates preexisting disorders, such as pneumonia or status asthmaticus, which present initially with hypoxemia without hypoventilation. Aministration of oxygen alone is not an appropriate treatment for hypercapnic respiratory failure and can result in the patient retaining even more carbon dioxide, especially in situations where the child has adapted to chronic hypercapnia and is relatively dependent on oxygen-sensitive peripheral chemoreceptors to maintain ventilatory drive. In addition to supplemental oxygen, therapies to reduce the load on the respiratory muscles and increase the level of alveolar ventilation should be instituted in children with type II respiratory failure. When to use NPPV for acute respiratory failure? When the cause of ARF is reversible, medical treatment works to maximize lung function and reverse the precipitating cause, whereas the goal of ventilatory support is to "gain time" by unloading respiratory muscles, increasing ventilation, and thus reducing dyspnea and respiratory rate and improving gas exchange. Two recent physiological studies have demonstrated these beneficial effects of NPPV in children with ARF [8, 9] . NPPV is increasingly used for treatment of ARF in children. Tables 1 and 2 summarize the studies reporting the effectiveness of NPPV in children with ARF of various etiologies [8, . However, the determinants of success of NPPV relate more prominently to the primary diagnosis as discussed below. Lower airway disease is a common cause of ARF. Asthma accounts for the largest percentage of this group, but infections, such as viral bronchiolitis, also are common and predominantly impact the small airways. Physicians caring for acutely ill children are regularly faced with this condition. Both non-invasive and invasive ventilation may be options when medical treatment fails to prevent respiratory failure. ETI and positive pressure ventilation in children with lower airway obstruction may increase bronchoconstriction, increase the risk of airway leakage, and has disadvantageous effects on circulation and cardiac output. Therefore, ETI should be avoided unless respiratory failure is imminent despite adequate institution of all available treatment measures. NPPV can be an attractive alternative to IMV for these patients. Clinical trials in children with acute lower respiratory airway obstruction have suggested that NPPV may improve symptoms and ventilation without significant adverse events and reduce the need for IMV [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] . NPPV theoretically improves the respiratory status of patients with lower respiratory airway obstruction by several mechanisms [37] . During acute bronchospastic episodes, patients have an increase in airway resistance and expiratory time constant. The combination of prolonged expiratory time constant and premature closure of inflamed airways during exhalation results in dynamic hyperinflation, which causes increased positive pressure in the alveoli at end-expiration (auto-PEEP). Because the alveolar pressure must be reduced to subatmospheric levels to initiate the next breath, this auto-PEEP increases the inspiratory load and induces respiratory muscle fatigue. The EPAP delivered by NPPV may help to decrease dynamic hyperinflation by maintaining small airway patency and may reduce the patient's work of breathing by decreasing the drop in alveolar pressure needed to initiate a breath. In addition, inspiratory support, i.e., IPAP delivered by NPPV, helps to support fatigued respiratory muscles, thereby improving dyspnea and gas exchange. Needleman et al., in a physiological study, found that the NPPV use in children with status asthmaticus was associated with a decrease in respiratory rate and fractional inspired time and an improvement of thoracoabdominal synchrony in 80% of patients [12] . A few clinical studies of small size (3-73 patients) reported the use of NPPV for treatment of status asthmaticus in children (Table 1 ) [10, 11, 13, 14] . NPPV was well tolerated with no major complications and was associated with an improvement of gas exchange and respiratory effort (Table 1) . Viral bronchiolitis, mainly due to respiratory syncytial virus, represents the largest cohort of children treated with NPPV [15] [16] [17] [18] [19] [20] . Use of NPPV in infant with severe bronchiolitis was associated with improved respiratory rate [15, 19] and PaCO 2 [16, 19, 20] , decreased work of breathing [17] , and ETI avoidance in 67-100% of patients (Table 1 ) [17, 18] . In children, dynamic upper airway obstruction can present as an acute life-threatening condition and leads to severe alveolar hypoventilation. In 2006, a survey of French PICU group found that 67% of pediatric intensivists applied frequently or systematically NPPV in the management of dynamic upper airway obstruction in children [38] . However, there is a paucity of literature on the use of NPPV in the acute setting of upper airway obstruction in children. NPPV was associated with a significant decrease in respiratory effort [21] and a sustained improvement in gas exchange [22] in children with dynamic upper airway obstruction (Table 1) . The main goals of NPPV in patients with parenchymal lung disease, such as pneumonia, acute lung injury (ALI), and acute respiratory distress syndrome (ARDS), are to improve oxygenation, to unload the respiratory muscles, and to relieve dyspnea. The first goal can usually be achieved by using EPAP to recruit and stabilize previously collapsed lung tissue [39] . Unloading of the respiratory muscles during NPPV with IPAP has been reported by L'Her et al. in adult patients with ALI [39] . The authors concluded that adding IPAP to EPAP may be indispensable in patients with ALI treated with NPPV [39] . Indeed, IPAP allows a better respiratory system muscle unloading, alveolar recruitment, oxygenation, and CO 2 washout improvement. Although NPPV seems disappointing in ARF owing to pneumonia in adult patients, with failure rates of up to 66% [40] , several noncontrolled trials have suggested that NPPV could improve symptoms and ventilation without significant adverse events and reduce the need for IMV in children with ARF due to pneumonia [22] [23] [24] [25] [26] [27] . Use of NPPV in this acute setting in children was associated with reduction in ETI rates ranging from 50-100% (Table 1) [23, 24] . The most challenging application of NPPV may be in patients with ARDS. Studies of NPPV for the treatment of ARDS in adult population have reported failure rates of 50-80% [40] . A meta-analysis of the topic in adult population concluded that NPPV was unlikely to have any significant benefit [41] . In children, the use of NPPV for the treatment of ARDS was associated with a failure rate of 78%, and 22% of them died (Table 1 ) [27] . Therefore, NPPV use in such a patient group is rarely justified. However, if a trial of NPPV is initiated, patients should be closely monitored and promptly intubated if their conditions deteriorate, so that inordinate delays in needed interventions are avoided. Acute chest syndrome (ACS) is one of the leading causes of death and hospitalization among patients with sickle cell disease [42] . Approximately 70% of patients (adults or children) with ACS are hypoxic [43] . Indeed, patients with sickle cell disease are prone to infarctive crises. Thoracic bone infarction (usually in the ribs) in such patients leads to pain, splinting, hypoventilation, and the clinical signs of ACS. In situ red blood cell sickling in the lung vasculature is possibly a consequence of hypoventilation with subsequent infarction of lung parenchyma. NPPV has been proposed as a therapeutic option for patients with ACS. By improving patient oxygenation, NPPV could prevent progression from painful crisis to ACS, and ultimately to ARDS. Three retrospective studies reported favorable outcomes in children with ACS treated with NPPV (Table 1) [22, 27, 28] . Postoperative pulmonary complications are a major cause of morbidity, mortality, prolonged hospital stay, and increased cost of care [44] . It has been reported that 5-10% of all surgical adult patients experience postoperative pulmonary complications [45] . Atelectasis, postoperative pneumonia, ARDS, and postoperative respiratory failure have all been classified as postoperative pulmonary complications. Postoperative respiratory failure is most commonly defined as the inability to be extubated 48 hours after surgery [46] , although some investigators have used 5 days [47] . NPPV has been successfully used to treat postoperative respiratory failure in both pediatric and adult patients. Compared with standard treatment, NPPV used after major abdominal surgery improved hypoxemia and reduced the need for ETI in adult population [48] . NPPV application in children with postoperative respiratory failure was associated with improved respiratory effort, gas exchange, oxygen saturation, and reduced the need for ETI (Table 1 ) [8, 24, 26, 27, 29, 30] . The need for reintubation after failed extubation is associated with increased morbidity and high mortality [49] . NPPV has been proposed as a means of "facilitating" weaning from IMV, and as a "curative" treatment for postextubation respiratory failure. Although several studies have shown the efficacy of NPPV in weaning from IMV in adult population [50] , its application for postextubation respiratory failure is not supported by randomized, controlled trials [51] . In children, two noncontrolled trials assessed the efficacy of NPPV in these settings: the application of NPPV as a means of "facilitating" ventilation weaning, and as "curative" treatment for postextubation respiratory failure was associated with success rates of 81-86% and 50-75%, respectively [31, 32] . ARF in immunocompromised patients most often results from infections, pulmonary localization of the primary disease, or even postchemotherapy cardiogenic pulmonary edema. Treatment of such patients often requires intubation and mechanical ventilation. Avoidance of the infectious complications associated with IMV is particularly attractive in these high-risk patients, in whom this could be devastating, if not fatal. Results of randomized, controlled trials have proven the beneficial effects of NPPV in immunocompromised adult patients [52, 53] . Some case series reported the use of NPPV in the treatment of respiratory failure in immunocompromised children (Table 2 ) [23, 27, [33] [34] [35] [36] . The likelihood of NPPV success in immunocompromised children seems to be related rather to the type of pulmonary disease: the ETI avoidance rates varied from 40% for ARDS to 100% for pneumonia ( Table 2) . Are there predictive factors of NPPV failure in children with ARF? It is not always apparent which patients will initially benefit from NPPV; some patients do not obtain adequate ventilation with NPPV. The NPPV failure rate may be fairly consistent for certain diseases, and NPPV failure eventually requires intubation. Inability to early identify patients who will fail NPPV can cause inappropriate delay of intubation, which can cause clinical deterioration and increase morbidity and mortality. Knowing the predictors of NPPV failure in patient with ARF is therefore crucial in deciding if and when to apply this ventilatory technique. Several authors have identified different predictive factors of NPPV failure in children with ARF: the results of studies are given in Table 3 [20, 24, 26, 27, 31, 54, 55] . The best predictive factors for the NPPV failure in ARF appear to be the level of FiO 2 and PaCO 2 on admission or within the first hours after starting NPPV (Table 3) . During recent years, there has been an increasing interest in the use of NPPV for children with ARF. There are some promising studies supporting its use in this acute setting. NPPV was well tolerated with rare major complications and was associated with improved gas exchange, decreased work of breathing, and decreased need for ETI. Both critical care ventilators and portable ventilators have been used for NPPV. However, the vast majority of the available knowledge in this acute setting results from noncontrolled trials and case series of small size. As such, many important issues, such as the identification of the patient, the right time for NPPV application, and the appropriate setting, are still lacking. Further randomized, controlled trials addressing these issues in children with ARF are needed to define better the patients who are likely to benefit from this alternative method of respiratory support. Also, the respective place of NPPV and high flow oxygen therapy in children with ARF due to different conditions has to be determined [56] . Najaf-Zadeh and Leclerc Annals of Intensive Care 2011, 1:15 http://www.annalsofintensivecare.com/content/1/1/15

Pandemic A/H1N1v influenza 2009 in hospitalized children: a multicenter Belgian survey

BACKGROUND: During the 2009 influenza A/H1N1v pandemic, children were identified as a specific "at risk" group. We conducted a multicentric study to describe pattern of influenza A/H1N1v infection among hospitalized children in Brussels, Belgium. METHODS: From July 1, 2009, to January 31, 2010, we collected epidemiological and clinical data of all proven (positive H1N1v PCR) and probable (positive influenza A antigen or culture) pediatric cases of influenza A/H1N1v infections, hospitalized in four tertiary centers. RESULTS: During the epidemic period, an excess of 18% of pediatric outpatients and emergency department visits was registered. 215 children were hospitalized with proven/probable influenza A/H1N1v infection. Median age was 31 months. 47% had ≥ 1 comorbid conditions. Febrile respiratory illness was the most common presentation. 36% presented with initial gastrointestinal symptoms and 10% with neurological manifestations. 34% had pneumonia. Only 24% of the patients received oseltamivir but 57% received antibiotics. 10% of children were admitted to PICU, seven of whom with ARDS. Case fatality-rate was 5/215 (2%), concerning only children suffering from chronic neurological disorders. Children over 2 years of age showed a higher propensity to be admitted to PICU (16% vs 1%, p = 0.002) and a higher mortality rate (4% vs 0%, p = 0.06). Infants less than 3 months old showed a milder course of infection, with few respiratory and neurological complications. CONCLUSION: Although influenza A/H1N1v infections were generally self-limited, pediatric burden of disease was significant. Compared to other countries experiencing different health care systems, our Belgian cohort was younger and received less frequently antiviral therapy; disease course and mortality were however similar.

On March 2009, in Mexico, a novel recombinant influenza strain (A/H1N1v) of swine origin was discovered as an infective agent in humans [1] . This new virus spread rapidly, first to USA and Canada, then all over the world, causing the "new 2009 influenza A/H1N1v pandemic" [2] . Worldwide, the burden of disease was significant and subsequent efforts from health care systems were required to face an overload of patient's consultations as well as to implement vaccination and surveillance programs. Although consequences of the pandemic were less dramatic than initially feared, the World Health Organization (WHO) estimated that the virus was responsible of at least 17700 deaths worldwide and the Centers for Disease Control and Prevention (CDC) reported 59 millions infected people in the USA [3] [4] [5] ). During this A/H1N1v flu wave, children and young adults were identified as a particular risk group. They presented a higher attack rate than older adults [6] and a greater mortality rate than previously observed with classical seasonal flu [7, 8] . Several reports on influenza A/H1N1v in pediatric settings have now been published [9] [10] [11] [12] , but information on clinical presentation and severity of infection in European children remains limited [13] [14] [15] . However, these data could be of the great interest to guide future recommendations for vaccination and antiviral therapy during forecoming flu seasons. Belgium experienced the influenza A/H1N1v epidemic from July 2009 to January 2010. Pandemic vaccine (adjuvanted Pandemrix ® ) was only available after the peak occurred in October and was given with priority to risk groups (defined as health care workers, pregnant women and chronically ill patients) [16] . According to our national surveillance system, around 214531 people were infected, 733000 could benefit from vaccination and 19 deaths were attributable to the virus [17] . In this context, we conducted a multicenter study analyzing influenza A/H1N1v pediatric cases hospitalized in four tertiary medical centers of Brussels, Belgium. Our study aimed to offer a comprehensive description of influenza A/H1N1v infection in children, in the light of other recently published data from countries experiencing different health care systems [9] [10] [11] [12] . In collaboration with infection control units and microbiology laboratories, we prospectively registered all proven and probable pediatric cases of influenza A/H1N1v infections hospitalized in four tertiary facilities of Brussels (Hôpital Universitaire des Enfants Reine Fabiola, Universitair Ziekenhuis of Brussels, Cliniques Universitaires Saint-Luc and Hôpital Saint-Pierre). These facilities totalize 406 pediatric beds, representing 80% of the total pediatric beds available in Brussels (about 1 million inhabitants in 2009). Moreover, three of the hospitals have a Pediatric Intensive Care Unit (PICU) where critically-ill children from other hospitals of Brussels and the surrounding areas are referred to (in total 32 PICU beds available). The study period extended from July 1, 2009, to January 31, 2010. Children were included if they were aged from 0 to 18 years, presented with clinical symptoms compatible with influenza (fever and/or respiratory signs/symptoms) and had either positive PCR results for influenza A/H1N1v (proven cases), or an antigen and/or a positive culture for influenza A (probable cases). The latter cases were included because virtually no other seasonal influenza A viruses were circulating in Belgium during the epidemic period (less than 0.4%, data from the Belgian National Institute of Public Health). Moreover, specific H1N1v PCR confirmation was no longer carried out routinely at the end of the epidemic, due to the high cost of this testing and the limited number of cases after December 2009. Data were collected retrospectively from patients' medical files using a standardized questionnaire. A pre-existing co-morbidity was defined as a chronic condition requiring long term medication or medical follow up. Co-morbidities were listed based upon CDC H1N1 flu guidelines http://www.CDC.gov/h1n1flu.htm and other recent publications [8] . Co-morbidities were not mutually exclusive, so that a child could participate in several categories. Fever was defined as a central temperature above 38°C elsius. Nosocomial infection was defined as a proven or probable case occurring after more than 48 hours of hospitalization. Respiratory samples collection included nasopharyngeal aspirates, nasopharyngeal flocked swabs (Copan Diagnostics, Corona, CA), throat flocked swabs (Copan Diagnostics, Corona, CA) and sputum. Antigen testing was assessed by immunochromatographic rapid antigen detection (RAT) in three of the four centers or also by direct immunofluorescence (DIF) technique (Argene SA, Verniolle, France) in one of them. Both DIF and immunochromatography use highly sensitive monoclonal antibodies directed against either influenza A or B nucleoprotein antigens [18] . RAT was performed using two different testing: the Coris Influ-A&B Respi-Strip (Coris Bioconcept, Gembloux, Belgium) and the Binax Now influenza A & B (Binax Inc., Inverness medical, Maine, USA). Direct antigen testing was unavailable in the fourth participating hospital, representing 26% of our cohort of patients. Viral culture was performed on the three following cell lines: Vero, MRC-5 and LLC-MK2 in two centers; and MDCK, Hep-2, MRC-5 and LLC-MK2 in a third hospital, as described elsewhere [19] . In the fourth center which used DIF and RAT for antigen detection, respiratory samples were not cultured as antigen detection was followed directly by real-time PCR. (This center represented 22% of the cohort). Biomolecular testing consisted, for three of the four centers, firstly in detection of influenza A virus by a home-made real-time RT-PCR (RT-PCR InflA) targeting the matrixprotein-coding gene and consequently by specific detection of the circulating pandemic variant using two monoplex real-time RT-PCR assays as described in the Centers of Disease Control (CDC) protocol: the SW InfA PCR (SWINE) and the SW H1 PCR (RT-PCR A/ H1N1) [20] . In the fourth center (26% of samples), a commercial available PCR kit was used directly for detecting the pandemic strain: "Swine influenza virus (sw H1N1) Real-time PCR" (Diagenode Diagnostics, Liège, Belgium). Statistical analyses were performed using Graph Pad Prism Software, Inc, 2003, San Diego, USA. Chi square or Fisher's exact tests were used to compare non continuous variables and Mann Whitney u-test was used to compare continuous variables. A two-tailed p-value less than 0.05 was considered as statistically significant. Approval of the Medical Ethics Committees of the four hospitals was obtained before starting the study. A code number was attributed to each child so that data collected remained strictly confidential. During the epidemic period, an excess of 18% (+10486) of pediatric outpatients and emergency department visits was registered, as compared with the mean measured over the 3 previous years during the same months. Figure 1 represents the evolution of H1N1v 2009 pediatric hospitalized cases over time in the four hospitals, with peak of the epidemic observed between the end of October and the beginning of November 2009. 215 children were hospitalized with proven or probable influenza A/ H1N1v infection; representing 2% of the total hospitals' admissions registered during the whole study period but 6% (191/3144) of those during the four weeks of the peak of the epidemic. Additionally, the PICU occupation rate by influenza A/H1N1v infected children was 3.5% over the whole study period and reached 8% during the peak of the epidemic. Among our cohort of 215 children, 57% were male. The median age of the patients was 31 months (range: < 1 to 208 months), with 19% of the children having less than 3 months of age ( Figure 2 ). As shown in Table 1 , 101/215 (47%) children presented with one or more underlying co-morbid condition, principally chronic lung diseases and neurological disorders. The median age of patients presenting co-morbidities was significantly higher than of those without (50 versus 14 months, p < 0.0001). The major clinical features, reasons for hospitalization and blood diagnostic results are summarized in Table 2 . The median duration of symptoms before admission was 2 days (IQR: 1-4 d). Presentation on admission consisted mainly in febrile respiratory illness, with a high prevalence of gastrointestinal symptoms, independently of age groups (Table 3) . Moreover, 21 children (10%) presented initially with neurological manifestations. Based upon initial clinical presentation, diagnosis of influenza A/H1N1v infection was suggested by clinicians in 56% of children (Table 2) . 74/215 (34%) children had chest X-ray confirmed pneumonia, associated in 6% with pleural effusion. Three patients had confirmed bacterial superinfection with Streptococcus pneumoniae (positive blood cultures). For five patients (2%), the influenza infection was hospital acquired. As shown in Table 3 , children aged less than 3 months old had a significantly lower rate of pneumonia and tended to have less neurological manifestations than older ones. For this age group, the main reason for hospitalization was surveillance of acute fever without focus. Initial clinical presentation was globally similar between patients with and without chronic co-morbidities, except for hypoxemia (Table 3) . Blood inflammatory profile on admission was variable, with 13% of the children presenting with leukopenia and 6% with thrombocytopenia ( Table 2) . Respiratory samples collected to diagnose influenza A/ H1N1v infection were nasopharyngeal aspirates and nasopharyngeal swabs in 58% and 39% of the patients, respectively. Compared to PCR (considered as the gold standard), the sensitivity of antigen by RAT was low, being only 29% (10/34) and 57% (47/83) using Coris RespiStrip and Binax Now testing, respectively (data not calculated for DIF method, as only performed on 11 samples). The sensitivity of culture was quite better at 76% (107/141). Among the 215 children, only 51 (24%) received oseltamivir (doses according to weight and age [21] ). For a large majority of them (37/47, data unavailable for three children), antiviral therapy was started directly on admission and was continued for five days (35/41, data unavailable for 9 patients). Rate of oseltamivir prescription reached 42% and 71% for children having pneumonia and for those requiring PICU admission, respectively. No significant related adverse events were reported. Oseltamivir was significantly more frequently prescribed among children older than 2 years compared to younger ones (Table 4 ) and among children with underlying disease (39/101 [39%]) versus 12/114 [11%], p < 0.0001). Additionally, 57% (123/215) of the patients were treated with antibiotics, for a median duration of seven days. In only 17 of them (14%), antibiotic therapy was discontinued after obtaining confirmation of influenza A/H1N1v infection (data unavailable for seven cases). Among children treated by antibiotics, 54% had a diagnosis of pneumonia. Finally, antibiotics were similarly used in all patients' age groups (Table 4) . Intensive Care 21/215 patients (10%) had to be admitted to PICU, mainly within 24 hours of admission. Among them, the prevalence of co-morbidity (62%) tended to be higher than observed among ward patients, with a predominance of neurological disorders ( Table 1 ). The median age of PICU children was 75 months (IQR 47-130). PICU admissions were significantly more frequent in children above two years of age (Table 4 ) and no infant less than three months old required intensive care. The major reason for PICU admission was respiratory failure subsequent to pneumonia (Table 3) . Seven patients (3% of the global cohort) presented an Acute Respiratory Distress Syndrome (ARDS) and three had pleural effusion. Two patients were admitted for surveillance because of severe underlying disease. None of the 21 children presented seizures or signs of viral encephalitis ( Table 3 ). The median duration of PICU stay was four days and ranged from 1 to 90 days. 13/21 (59%) children received respiratory support with non invasive ventilation (NIV); eight (38%) required mechanical invasive ventilation for a median duration of six days (range 1 to 45). One previously healthy child presenting with severe ARDS followed by cardiac-respiratory arrest had to undergo Extra Corporeal Membrane Oxygenation (ECMO) support during 12 days, but survived with mild respiratory sequelae. The median duration of hospitalization was three days (IQR: 2-6 d). This result was unaffected by the patients' age (Table 4) . Parental request/non compliance 2 (1) Legend: *IQR = interquartile range; **PMN = polymorphonuclear leukocytes; ***CRP = C-reactive protein; 1 201 children with blood results available on admission; 2 defined as Hb oxygen saturation less than 94%; 3 transcutaneous saturation The case-fatality rate among the global cohort was 2% (5/215). The five deaths were directly attributable to influenza A/H1N1v with or without bacterial superinfection and occurred in children with co-morbidities who would otherwise have died from their underlying disease. These underlying diseases consisted in neurological disorders from various etiologies (extensive central nervous system glioma, polymalformative syndrome, Hurler syndrome, cerebral palsy and severe encephalopathy with pontocerebellar hypoplasy). For all of them, severe pneumonia was notified, associated for three children with ARDS. All deaths concerned children aged more than two years old ( Table 4 ). Four of the five children had received oseltamivir within 48 hours of clinical symptoms. Three additional patients (1%) were cured from influenza A/H1N1v infection but still suffered from sequelae at the end of the study (two had bronchiectasis with emphysema and one a pulmonary restrictive syndrome needing tracheotomy and NIV support at home). Even though consequences were less dramatic than initially feared, the 2009 influenza A/H1N1v pandemic has caused a significant burden of disease worldwide, especially in the pediatric population [6] [7] [8] . Higher attack rate was observed among children, causing important overload in outpatient and emergency departments [11] as well as in PICU [22] . Unfortunately, our study was not designed to assess the epidemiological impact of influenza A/H1N1v infection over the whole pediatric population of Brussels. Moreover, by selecting only laboratory confirmed infections, we underestimated the number of hospitalized cases, especially since diagnosis confirmation by PCR was no more routinely performed at the end of the epidemic. However, we were able to notice an important pediatric burden of disease in Brussels, as illustrated by an increased rate of outpatients visits of 18% during the epidemic period compared to the three previous years and a high rate of PICU occupation by influenza A/H1N1v infected patients (8% during the peak of epidemic). It would have been of interest to compare the rate of hospitalization related to H1N1 with those registered for seasonal flu during the 3 previous years but these data were unfortunately not available. This study offers a comprehensive description of influenza A/H1N1v pattern of infection among Belgian hospitalized children, in the light of recent publications from other continents. Consistently with these reports [9] [10] [11] [12] 23, 24] , co-morbidities were highly prevalent among influenza A/H1N1v infected hospitalized children (47%). The co-morbidities were not different from those observed during seasonal flu. As described by others [7, 9, 22] , the presence of at least one co-morbidity was significantly more frequent in children of more than two years of age and constituted a risk factor for severity of disease, in terms of PICU admissions and casefatality rate. Furthermore, influenza A/H1N1v illness course differed according to patients' age groups. Indeed, children less than two years of age (46% of the cohort), and especially those less than three months, presented milder patterns of infection and were often hospitalized only for observation of fever without focus. 86% of PICU admissions and all deaths occurred in children over two years of age (with 80% of deaths among children > five years old). Although this issue is conflicting in the medical literature [7, 10] , similar findings have been reported in a large series by investigators from the CDC [25] . This observation differs from what is seen during seasonal flu, where young children and especially infants presented a higher mortality-rate compared to older ones [26, 27] . Nevertheless, during this pandemic wave as well as during previous flu seasons, the highest rate of hospitalization was generally reported among young age groups [8, 9, 11, 24] . This was particularly true in our series, as reflected by a median of age of 31 months, which was even younger than among American and Israeli hospitalized children (median age ranging from four to six years) [9, 10] . If unexplained by the severity of infection, this finding probably illustrates differences in clinical practices and hospitalization policies. In Belgium, the National Healthcare System renders the access to inpatients pediatric facilities easy, so that hospitalization of infants presenting with fever without focus, especially those younger than 3 months of age, is largely recommended and not restricted to the most severe cases as in other countries [28, 29] . Although we focused on hospitalized cases, influenza A/H1N1v episodes were mainly self-limited, consisting of febrile respiratory illness and requiring short duration of hospitalization (median 3 days) with or without oxygen supplementation. Initial clinical features did not differ from seasonal flu [30] , except for the higher proportion of children presenting with gastrointestinal manifestations, as described in previous studies [12, 23, 24] . This involvement of the gastrointestinal tract could be subsequent to a high rate of influenza A/ H1N1v virus replication [31] . Neurological manifestations were also frequent (10% of children) but in contrast with other reports [9, 22] were not correlated with PICU admission or fatality. Finally, more than one third of the whole cohort and 71% of PICU patients had pneumonia confirmed on chest X-rays. Even though only 3 (1%) children had evidence of bacteremia (all due to S. pneumoniae), it seems very likely that a significant proportion of pneumonia, especially those with lobar infiltrates, were associated with bacterial super-infections. According to some series, bacterial super-infections after influenza A/H1N1v episodes were found in about 4% of hospitalized children [11, 12, 23] but reached 20 to 38% among fatal cases [7, 25, 32] . Considering the low rate of positive blood cultures (BC) in pediatric bacterial community-acquired pneumonia (2.5 to 5%) [33, 34] , these published rates as well as our data likely constitute an under-estimation, as bacterial pneumonia diagnosis relied on positive cultures from sterile sites or autopsies and as part of children had received antibiotics prior to microbiological documentation. Finally, among our whole cohort, no necrotizing pneumonia, empyema or sepsis due to S. aureus or group A Streptococcus were reported, even though those pathogens are frequently involved in other series [9, 25] . Surprisingly, a majority of children were treated by antibiotics, even after the diagnosis of influenza A/ H1N1v infection was obtained. As mentioned above, confirming bacterial pulmonary super-infection after influenza illness is challenging and diagnostic relies more on clinical presumption and unspecific blood results [35] . However, this could only partly explain the high rate of antibiotics use, as only 54% of those children treated by antibiotics presented pulmonary infiltrates. Rate of antibiotics prescriptions was uniform among all age groups and was also high in other pediatric studies [10] [11] [12] . The exact reasons sustaining this practice remain unknown but should be worth to investigate in further prospective studies. Contrastingly, our study showed a particularly low percentage of oseltamivir prescriptions. Indeed, only 24% of the children were treated compared with 45 to 84% in other similar studies [8, 9, 11, 12] . In our four centers, the use of oseltamivir was significantly higher in children above 2 years of age and/or suffering from comorbidities. This more "watch and wait" practice was in line with the restrictive national recommendations issued for oseltamivir pediatric use during the 2009 pandemic wave, which suggested cautious prescription under one year of age, regarding the absence of safety data among infants [16] . Moreover, in Belgium, prescription of antiviral drugs during seasonal flu is very limited and kept for management of severe diseases or immuno-compromised patients [36] . Obviously, the limitations associated with the retrospective design do not allow us to conclude on treatment efficacy. It is however interesting to note that, although oseltamivir was scarcely used, fatality rate and PICU admissions were comparable to the other above-mentioned reports [9, 10, 12] . In Argentina [8] , the case-fatality rate of influenza A/ H1N1v infected children was 5%, with a global pediatric mortality rate 10 times greater compared to previous flu seasons. National surveys in the United Kingdom [7] and U.S [23] reported also a higher influenza related mortality rate during the pandemic influenza A/H1N1v than observed with seasonal flu. However, consequences in the Northern hemisphere were less dramatic than anticipated. Studies from these countries reported case-fatality rate among hospitalized children ranging from 0.6 to 3% [9] [10] [11] [12] , similar to our findings (2%). As previously hypothesized [10] , these North/South differences in patients' outcome could partly be explained by an easier access to the health care system in Europe, Israel and North America. In Israel [10] , as well as in our series, the median duration of symptoms before hospitalization was only 2 days compared to 4 days in Argentina [8] . On another hand, 2 patterns of influenza A/H1N1v related deaths have been described [7] : those occurring after several days of hospitalization in chronically ill patients (80%), in contrast to those observed after acute evolution of viral infection in previously healthy children (20%). Despite a small number of cases, a similar profile seemed to happen in our series, as the five children who died suffered from chronic neurological disorders and one previously healthy child presented a fulminant viral infection causing cardio-respiratory arrest and requiring nine days of ECMO support to be cured. Although influenza A/H1N1v infections were globally self-limited, pediatric burden of disease was significant. Children of more than 2 years old and/or suffering from chronic co-morbidities were shown at higher risk of severe infection. Compared to other countries experiencing different health care systems, our Belgian cohort was younger and received less frequently antiviral therapy; disease course and mortality were however similar.

A Quantitative Method for the Specific Assessment of Caspase-6 Activity in Cell Culture

Aberrant activation of caspase-6 has recently emerged as a major contributor to the pathogeneses of neurodegenerative disorders such as Alzheimer's and Huntington disease. Commercially available assays to measure caspase-6 activity commonly use the VEID peptide as a substrate. However these methods are not well suited to specifically assess caspase-6 activity in the presence of other, confounding protease activities, as often encountered in cell and tissue samples. Here we report the development of a method that overcomes this limitation by using a protein substrate, lamin A, which is highly specific for caspase-6 cleavage at amino acid 230. Using a neo-epitope antibody against cleaved lamin A, we developed an electrochemiluminescence-based ELISA assay that is suitable to specifically detect and quantify caspase-6 activity in highly apoptotic cell extracts. The method is more sensitive than VEID-based assays and can be adapted to a high-content imaging platform for high-throughput screening. This method should be useful to screen for and characterize caspase-6 inhibitor compounds and other interventions to decrease intracellular caspase-6 activity for applications in neurodegenerative disorders.

Proteases of the caspase family are known as important mediators of apoptosis and have been commonly subdivided based on their roles in apoptosis or inflammation (apoptotic initiator, apoptotic executioner or inflammatory caspases). This definition however has become somewhat inaccurate as an increasing number of non-apoptotic roles for both initiator and executioner caspases have been identified that mediate cell differentiation, maturation and signaling events [1] . Caspases can further be distinguished based on their inherent differences in caspase substrate preference that are defined by the shape and electrostatic potential of the active site cleft [2] . Using positional scanning of peptide libraries, consensus recognition sequences have been proposed for each caspase and have led to the development of peptide substrates as well as inhibitors that typically consist of 4 amino acids (i.e. DEVD for caspase-3), followed by a fluorescent tag such as Afc (7-amino-4-trifluoro methylcoumarin) for a substrate or a 'warhead' such as fmk (fluoromethylketone) that covalently binds the enzyme for an inhibitor. These reagents are useful to investigate caspases that constitute the majority of caspase-like activity in a sample, as it may be assumed for active caspase-3 in highly apoptotic extracts [3] . However, with Km/kcat ratio differences of less than 10 fold for many widely used peptide substrates [4] , these reagents are not particularly useful for investigating the activity of a caspase present at lower concentrations in cell culture and tissue samples. In particular in developmental or signalling processes that do not involve cell death, intracellular caspase activity is likely under tight control by endogenous caspase inhibitors or the proteasome [5, 6] and the resulting low levels of activity are difficult to detect with peptide substrates. In biologic protease substrates, additional factors outside the 4 amino acid recognition site can influence the selectivity and efficiency of proteolytic cleavage. For caspases, it has been shown that the amino acid residue directly after the scissile bond (P19) is an important determinant of cleavage, since charged or bulky residues are not well tolerated [7] . Furthermore, domains far away from the cleavage site can mediate the interaction between substrate and protease (exosites), and although such interactions have not yet been shown for proteases of the caspase family, the high variability of cleavage site motifs in natural caspase substrates argues in favour of the presence of exosites. Known substrates for caspase-6 show a particularly high variability in their recognition sequences [8] , with cleavage sites other than I/D/E/L/T/V, E/ D/Q, X, D found in substrates such as the presenilins (ENDD, [9] ), huntingtin (IVLD, [10] ), DNA Topoisomerase I (PEDD, [11] ), AP-2 alpha (DRHD, [12] ), Periplakin (TVAD, [13] ), FAK (VSWD, [14] ) and TGEV (VVPD, [15] ). Caspase-6 has garnered much attention recently since it has been shown that it is involved in the developmental pruning of axons [16, 17] , and it has been suggested that similar pathways might erroneously be activated in neurodegenerative disorders such as Alzheimer's (AD) and Huntington disease (HD) [16, 18] . The presence of activated caspase-6 and cleavage of caspase-6 substrates is indeed a hallmark of AD, HD and cerebral ischemia, and has been shown in a number of different animal models and patient brain tissue [18, 19, 20, 21, 22] . To assess caspase-6 activity in cell and tissue samples, peptide substrates or inhibitors need to be titrated accurately to yield meaningful results, since the peptide substrate commonly used to assess caspase-6 activity, VEID, can be cleaved by other caspases as well as the proteasome when used at too high concentrations [23, 24] . In addition, even low concentrations of a VEID substrate can lead to inaccurate results if the relative amount of other proteolytic activity in the sample is significantly higher than that of caspase-6 due to small differences in kcat/Km [4] . A recently developed method using the biotinylated caspase inhibitor zVAD is significantly more specific, but to achieve this specificity, immunoprecipitation and subsequent detection of the active caspase by Western blotting is required [23] . For a more quantitative way of assessing caspase-6-specific activity, we have developed a novel method based on the cleavage of lamin A [24] . Here we show that this method is sensitive, specific and allows for the detection of caspase-6 activity in complex samples. The assay quantitatively measures caspase-6 activity and can be useful for investigating caspase-6 biology as well as the screening and intracellular efficacy assessment of caspase-6 inhibitors. In order to develop an activity assay that is more specific for caspase-6 than available methods relying on the cleavage of the VEID peptide, we decided to investigate the cleavage of known caspase-6 substrates. To be useful for the measurement of caspase-6 activity, a substrate would ideally be as specific as possible. In particular it should not be cleaved by other proteases activated during apoptosis and be cut by caspase-6 in detectable quantities. The protein substrate should be easy to purify and thus available in larger amounts, and existing neo-epitope antibodies allowing for the specific detection of the cleavage fragment would be an advantage. The nuclear lamin proteins were among the first caspase-6 substrates characterized [25] , and knockdown studies strongly suggest that their cleavage during apoptosis is highly dependent on the presence of caspase-6 [24] . Furthermore, purified lamin A as well as a variety of neo-epitope antibodies against the cleavage fragments are readily available, and high stability of the cleavage fragments without further degradation during the apoptotic process has been described [26] , which should facilitate their detection. The VEID peptide sequence commonly used in caspase-6 substrates and inhibitors is derived from the lamin cleavage site, but since protein context is known to change the kinetics and specificity of proteolytic events, we decided to compare the cleavage of the VEID peptide to that of the full-length lamin A protein. The lamin A protein is a more specific substrate than its VEID peptide We therefore decided to investigate whether lamin A is a specific substrate for caspase-6. In particular, we were interested in the ability of other executioner caspases to cleave lamin A, since these are likely to show high activity in apoptotic extracts where accurate quantification of caspase-6 activity will be of interest. We first subjected the Ac-VEID-AFC substrate to digestion by caspases -3, -6 and -7, respectively. The amount of active enzyme used in each reaction was normalized to the concentration of active sites in the sample as determined by titration against the irreversible inhibitor zVAD-fmk (Fig. S1 ). As expected, the VEID peptide substrate, although it is best processed by caspase-6, still shows significant cleavage by caspase-3 and -7 (Fig. 1A ) at high substrate concentrations (100 mM). Lower substrate concentrations can be used to achieve greater specificity, however, the difference in kcat/Km for VEID is less than 3 fold between caspase -3 and -6 [4] . This results in a small concentration window that can be exploited to achieve selectivity for caspase-6, if the same amounts of active caspase-3 and -6 are present. However, this window will be lost if the sample contains a higher concentration of active caspase-3 than caspase-6. As shown in Figure 1B , an 8 fold molar excess of caspase-3 over caspase-6 is enough to lead to a significantly higher signal from the non-specific caspase-3, making VEID-based assays problematic for the use in apoptotic samples or other cell and tissue lysates with high levels of active caspase-3. Lamin A, on the other hand, only showed proteolytic processing when incubated with recombinant active caspase-6, not with the corresponding amounts of caspases -3 and -7 at concentrations up to 300 mM (Fig. 1C) . We therefore focussed on lamin A as a substrate for the development of a specific caspase-6 activity assay. To confirm the selectivity of lamin cleavage by caspase-6, we made use of mouse embryonic fibroblast (MEF) cells generated from wild-type and caspase-62/2 (C6wt and C6ko) mice [27] . These cells only differ in their expression of caspase-6 ( Fig. 2A) , making them an ideal system to study the specificity of caspase-6 substrates. Apoptosis was induced by the addition of 50 nM staurosporine to the culture medium, and the activation of caspase-6 was monitored over time by assessing the cleavage of endogenous lamin A via Western blotting (Fig. 2B ) or with the Ac-VEID-AFC peptide substrate (Fig. 2F ). The antibody used to detect full length lamin A at 70 kDa cross-reacts with the closely related lamin C protein (60 kDa, [28] ), and the N-terminal fragments of both cleaved lamin A and C are detected at the same size of 28 kDa (Fig. 2B+E) . Furthermore, using a cell line that does not express caspase-3 but expresses caspase-6 ( Fig. 2C ), we found that the lamin proteins are still processed after the induction of apoptosis with camptothecin ( Fig. 2D ), confirming the requirement of caspase-6 but not caspase-3 for this process. An increase in VEID cleavage over time was observed for both C6wt and C6ko cell lines (Fig. 2F ). For C6wt cells, significant increases over baseline were observed at all time points, whereas the C6ko cells show lower VEID processing initially with a nonsignificant increase in the first 2 h of treatment. However, at later timepoints C6ko cells exhibit similar levels of VEID cleavage as C6wt with a highly significant increase over non-treated controls (Fig. 2F) , suggesting that other proteases make up for a large proportion of the VEID cleavage activity in apoptotic extracts. These results indicate that the lamin protein substrates are much more specific for cleavage by caspase-6 than the VEID peptide. Development of an electrochemiluminescence-based ELISA method for the quantitative assessment of cleaved lamin A In order to accurately quantitate the amount of cleaved lamin A generated by caspase-6, we turned to an ELISA-based assay format using the MesoscaleH platform, which allows for a fast and sensitive detection with minimal background using electrochemi-luminescence [29] . To determine whether cleavage of lamin A is more sensitive than cleavage of the VEID peptide substrate in detecting low levels of active caspase-6, we incubated different concentrations of the enzyme in parallel either with Ac-VEID-AFC or purified lamin A protein and determined the amount of cleavage after 30 min by measuring the fluorescence in the sample (for Ac-VEID-AFC) or by subjecting the sample to an ELISA using the lamin A neo-epitope antibody (for lamin A cleavage). Comparison of the results showed that both assays perform at least equally well in detecting active caspase-6 concentrations down to 10 nM, with the lamin cleavage assay showing a linear concentration-response relationship down to the lowest caspase concentrations tested (Fig. 3A) . Signal-to-noise ratios were at or above 3 for the ELISA assay for as low as 10 nM caspase-6, whereas the signal-to-noise dropped below 3 for the VEID-based assay at this concentration (Fig. 3B ). Using a peptide inhibitor for caspase-6, VEID-CHO, both assays furthermore arrived at a similar IC 50 value (lamin-based ELISA: 64620 nM, VEID-based assay: 5666 nM), indicating that the ELISA method is suitable to assess the inhibition of caspase-6 by small molecules or peptides. Overall, our method shows a slight increase in sensitivity over commonly used VEID cleavage methods and is able to reliably detect caspase-6 concentrations down to 10 nM. Next, we wanted to assess whether the lamin cleavage assay shows higher specificity than the VEID-based system for caspase-6 over other proteases that are activated after induction of apoptosis. To this end, we tested lysates derived from staurosporine-stressed C6wt and C6ko MEFs with our newly developed ELISA and found a linear increase in signal over time in wt samples, whereas the signal from C6ko samples remained stable at background levels that were similar to the background seen in untreated C6wt (Fig. 3D ). This indicates that the amount of intracellularly cleaved, Figure 1 . The lamin A protein is a more specific substrate than its VEID peptide. A: 100 mM VEID-Afc was incubated with different amounts of caspase -3, -6 or -7 for 1 h at 37uC. Fluorescence generated by cleavage was monitored over time, and the initial, linear portion of the curve was used to calculate the reaction velocity. VEID is preferentially cleaved by caspase-6, but also cross-reacts with caspases -3 and -7 at higher concentrations. Error bars are the SEM of N$3 of 3 independent experiments. B: 5 mM VEID-Afc was incubated with 0.5 mM caspase-6 or different amounts of caspase-3 for 1 h at 37uC. The reaction velocity was calculated as in A. Even at this low concentration of VEID-Afc, the peptide substrate can be cleaved by caspase-3, and an 8 fold higher molar concentration of caspase-3 than caspase-6 results in a higher signal for VEID cleavage by caspase-3 than caspase-6. Error bars are the SEM of N = 3 independent experiments, statistical significance was assessed by 1-way ANOVA and posthoc Dunnett comparisons: *** p,0.0001. C: Pure lamin A protein was incubated with different amounts of caspase -3, -6 or -7 for 30 min at 37uC. Samples were separated by SDS-PAGE and both fragments of cleaved lamin A was detected by Western blotting with antibodies #2031 (full-length lamin A and N-terminal fragment) and #2032 (C-terminal fragment). No cleavage was observed with caspases -3 or -7, while caspase-6 generated lamin A fragments in a dose-dependent manner. A representative image of 3 independent experiments is shown. doi:10.1371/journal.pone.0027680.g001 endogenous lamin A protein is a highly specific readout for the quantification of caspase-6 activated during staurosporine-induced apoptosis. Kinetic measurements of either Ac-VEID-AFC or lamin A cleavage by fluorescence or our newly developed ELISA allowed comparison of the kinetic parameters of caspase-6 for the two substrates. We find that the full-length lamin A protein has a more than 1000fold lower Km than its cleavage site peptide VEID (Table 1) . Although the kcat value is also decreased, the kcat/Km is still more than 10fold higher for lamin A (Table 1 ). This indicates that the binding between lamin A and caspase-6 is tighter, which might be mediated by domains outside the cleavage site. Such binding sites could also be responsible for the observed specificity of caspase-6 for the full-length protein substrate. The Km and kcat values for VEID are furthermore in good agreement with previously reported data [4] . Caspase-6 is postulated to be involved in neuronal degeneration and apoptosis [16, 17, 18, 20, 22, 30] , and detection of its activity in neuronal cultures is therefore of paramount interest. Previous studies have frequently used TUNEL assays as a measure of apoptosis, looked at VEIDase activity or the presence of the active caspase-6 fragment by Western blotting or immunostaining and assessed the presence of cleaved caspase-6 substrates [8, 18, 20, 21, 22] . The most specific methods to assess the effect of drug treatments on caspase-6 activity developed so far use antibodies against the active caspase-6 fragment, levels of which Figure 2 . VEID, but not lamin A+C, is cleaved in the absence of caspase-6. A: Caspase-6 protein (full-length, 32 kDa) is detected in MEFs generated from C6wt, but not C6ko mice. B: Endogenous lamin A protein (70 kDa) is cleaved in wt, but not C6ko MEFs after staurosporine stress for 4 h or longer. The antibody cross-reacts with full-length lamin C (60 kDa), and the cleaved band at 28 kDa has the same size for both lamin A+C (lower panel). C: MCF-7 cells express caspase-6, but not caspase-3 protein, whereas both C6wt and C6ko MEFs contain both caspases. hu: human, m: mouse, ns: non-specific band. D: MCF-7 cells were stressed with 5 mM camptothecin for different amounts of time and the cleavage of endogenous lamin A and C proteins was monitored by Western blotting with antibodies antibodies #2031 (full-length lamin A+C) and #2032 (C-terminal fragments). E: Schematic representation of lamin A and C and the caspase-6 cleavage site at AA 230. The N-terminal fragments generated by caspase-6 cleavage (red) have the same size (28 kDa) for both lamin A+C. F: C6wt or C6ko MEFs were stressed with 50 nM staurosporine for different amounts of time, lysates were generated and analyzed for cleavage of VEID-Afc. C6wt cells show a significant increase in fluorescence at each timepoint, the fluorescence signal obtained from C6ko lysates only reach a statistically significant difference from baseline after 4 h. would not necessarily change upon inhibitor treatment, or immunoprecipitation of active caspases with biotinylated zVADfmk, which depends on efficient precipitation and Western blotting for quantification [23] . Our newly developed ELISA method, however, could not be directly applied to neuronal cultures, since lamin A is not expressed in embryonic mouse brain [31] . Neurons in early stages of development up until postnatal day 5 express lamins of the B1 and B2 subtypes, which differ from the lamin A and C sequence at the caspase cleavage site (VEVD in B-type lamins and VEID in lamins A and C (Fig. 4A) [24, 31] ). Furthermore, the cleavage of lamins B1 and B2 at this site is not specific for caspase-6, as has been shown in a caspase-6-deficient cell line [24] . In agreement with these findings, we observed cleavage of lamin B1 in primary cortical neurons derived from both C6wt and C6ko mice at embryonic day 16.5 ( Fig. 4B) , while lamins A and C were not detected (data not shown). To overcome this obstacle, we decided to spike C6wt and C6ko neuronal lysates with purified lamin A protein, and after incubation at 37uC to allow for its cleavage by endogenous caspase-6 activity, we subjected the samples to our ELISA assay. We detect a significant increase in cleavage of lamin A protein when the incubation was performed in the presence of extracts derived from camptothecin-stressed C6wt neurons, whereas no increase in cleaved lamin A signal was observed in the presence of camptothecin-stressed C6ko neuronal extracts (Fig. 4C) . The ELISA method is therefore also suitable to detect caspase-6 activity in samples that do not contain endogenous lamin A or C protein. Detection of caspase-6 activity in the absence of endogenous lamin cleavage Caspase-6 activity can be localized to the nucleus, which is commonly associated with cell death and the cleavage of nuclear substrates such as lamin A, whereas a cytoplasmic localization of active caspase-6 as it is the case in neurodegeneration does not result in immediate apoptosis [5, 21, 32] . We therefore decided to test whether spiking of cell extracts with pure lamin A protein can detect active caspase-6 at early timepoints, before its translocation to the nucleus and cleavage of endogenous lamin A. To this end we transiently transfected COS7 cells with full-length human caspase-6, a system in which the enzyme auto-activates in the cytosol before translocating to the nucleus [32] . After transfection, the active form of caspase-6 becomes detectable by Western blotting at the 9 h timepoint (Fig. 5A) . After 24 h of incubation, low levels of endogenous lamin A cleavage were observed with the ELISA method, indicating that at this time point active caspase-6 is present in the nucleus (Fig. 5B) . However, when cell lysates were supplemented with purified lamin A protein and endogenous, active caspase-6 was allowed to cleave the spiked protein, activity could already be detected after 9 h and increased dramatically at later time points (Fig. 5B ), in agreement with the Western blot data (Fig. 5A ). This indicates that the spiking method can detect caspase-6 activity before endogenous lamin is cleaved and is thus suitable to detect non-nuclear caspase-6 activity. Although our Mesoscale ELISA method shows significant advantages over commonly used, VEID-based assay systems to measure caspase-6 activity, it still requires manual cell lysis and protein quantification steps that are not easily amenable to highthroughput screening campaigns. We therefore turned to a highcontent imaging platform that allows automated liquid handling as well as immunofluorescence imaging and quantification. Using a primary antibody specific for caspase-6 cleaved lamin A, we found increased perinuclear staining in C6wt cells after induction of cell death by camptothecin (Fig. 6A) . The observed blebbing of the nuclear membrane is consistent with the breakdown of the nuclear lamina during apoptosis. No staining for cleaved lamin protein was observed in non-stressed C6wt, non-stressed C6ko or camptothecin-stressed C6ko cells (Fig. 6A ). Using nuclear DAPI staining as a reference point, we quantified the perinuclear immunofluorescence in a ring around the nucleus (Fig. 6A, insets) , and found a 2fold increase in staining intensity in camptothecin-stressed versus non-stressed wt MEFs (Fig. 6B) . Only background signal was observed in C6ko MEFs even after camptothecin treatment, indicating that the detection of cleaved lamin A by immunofluorescence is also specific to caspase-6 (Fig. 6B) . This method is ideally suited for the high-throughput intracellular testing of modulators of caspase-6 activity in in the presence of confounding proteolytic activity, i.e. that of caspase-3. The study of the role of caspases during apoptosis, but also during non-apoptotic developmental and signalling processes, is hampered by the lack of specific assays to measure the activity of single members of the caspase family. Neo-epitope antibodies against the active forms of caspases are available and can be used in immunohistochemical and Western blotting applications. However, the presence of active caspase fragments does not necessarily correlate with proteolytic activity, since the proteases could be bound to either endogenous or exogenous inhibitors, which is especially problematic in high-throughput screening campaigns aiming to identify caspase inhibitors. Another method using immunoprecipitation with biotinylated zVAD, on the other hand, is specific to active caspases, but quantitation relies on the efficacy of immunoprecipitation and Western blotting [23] . To be able to accurately quantify the activity of caspase-6 in cell culture, we therefore developed novel activity assays based on the cleavage of the caspase-6 specific substrate protein lamin A [24] . Here, we confirm the specificity of lamin A cleavage by caspase-6 using C6wt and C6ko MEFs that were stressed with staurosporine to undergo apoptosis. Although the antibodies we used in this study cross-react with cleaved lamin C, specificity for caspase-6 cleavage is maintained, since both proteins arise from the same gene by alternative splicing, contain the same caspase-6 cleavage site and only differ in their C-termini with lamin A being 10 kDa longer (Fig. 2D) [28, 33] . The cleaved fragment was detected in C6wt, but not in C6ko MEFs, and we developed an electrochemiluminescence-based ELISA method to accurately quantify cleaved lamin using the Mesoscale platform. The new method has an improved detection limit and signal-noise ratio over commercially available caspase-6 activity assays using the VEID peptide substrate. Furthermore, our assay is specific to caspase-6 and thus superior for the detection of caspase-6 activity in complex samples with a variety of different proteolytic activities such as cell lysates. Through spiking with purified lamin A protein, the assay can be applied to samples lacking endogenous lamin A and C such as embryonic primary neurons. Furthermore, the method is amenable to samples where caspase-6 activity does not lead to the cleavage of the endogenous nuclear lamins and is therefore not necessarily associated with apoptosis. The specificity is due to the use of a protein instead of a peptide substrate, and even though the VEID sequence is derived from the caspase-6 cleavage site in lamin A (aa227-230), we show that the Kcat/Km value for lamin A is more than 10fold higher than for VEID, indicating that lamin A is a better substrate for caspase-6. A similar effect has been shown for proteases involved in blood Figure 6 . Caspase-6 activity can be quantified by immunofluorescent staining for cleaved lamin A. A: C6wt and C6ko MEFs were stressed with camptothecin, stained with an antibody against cleaved lamin A and analyzed on an automated imaging platform. Nuclei were counterstained with DAPI, and identified by the software (blue circles). Debris (red circles) was not analyzed. The signal intensity in the cleaved lamin A channel was quantified in a ring around the nucleus (green circles). B: Quantitation of the perinuclear staining from cleaved lamin A is graphed. Untreated C6wt MEFs were normalized to 100% and the fold change in stressed C6wt, stressed and non-stressed C6ko MEFs are compared. Error bars are the SEM of N$3 of 3 independent experiments. Statistical significance was assessed by 2way ANOVA and post-hoc Bonferroni comparisons: *** p,0.0001. doi:10.1371/journal.pone.0027680.g006 coagulation [34, 35] , where substrates are bound outside the active site of the protease (exosite). There has been speculation about the presence of exosites in caspases [36] , but their existence has not been proven conclusively yet. The example of exploiting protein substrate specificity for the development of an activity assay could be used for the development of more specific assays for other members of the caspase family, such as p23 for caspase-7 [3] . Caspase-6 has recently emerged as an important player in neuronal dysfunction and degeneration and its activation has been linked to several neurodegenerative conditions such as AD, HD and stroke [16, 17, 18, 19, 20, 21, 22, 30, 37] . Direct inhibition of the enzyme as well as targeting of other players in the activation pathway might therefore be beneficial in these disorders. We show here that the cleavage of lamin A can be assessed in a highthroughput setting using a high-content imaging platform, making it an ideal intracellular readout to test interventions that decrease caspase-6 activity. All experiments were carried out in accordance with protocols (Animal protocol A07-0106) approved by the UBC Committee on Animal Care and the Canadian Council on Animal Care. Ac-VEID-Afc, Ac-DEVD-Afc, Ac-VEID-CHO, zVAD-fmk and active caspase-3, -6 and -7 enzymes were purchased from Enzo Biosciences. Lamin A antibodies were from Cell Signaling Technology: cleaved lamin A and total lamin A/C for Western blotting (cat. nos. 2031 and 2032), cleaved lamin A for Mesoscale ELISA (cat. no. 2036) and immunofluorescence staining (cat. no. 2036). Pure lamin A protein and lamin B1 antibody were from Abcam (cat. nos. ab8982 and ab83472), antibody against fulllength caspase-6 was from Cell Signaling Technology (cat. no. 9762), antibody against caspase-3 was from Cell Signaling Technology (cat. no. 9662), antibody against actin was from Chemicon (cat. no. MAB1501R). Camptothecin and staurosporine were from Sigma, cell culture reagents were from Gibco. The amounts of enzyme indicated in the figures (5-500 nM) were incubated with 100 mM Ac-VEID-AFC at 37uC in caspase cleavage buffer (50 mM HEPES pH 7.4, 100 mM NaCl, 0.1% CHAPS, 1 mM EDTA, 10% glycerol, 10 mM DTT) for 1 h in a black 96well plate (Nunc). Fluorescence was measured every 5 min with excitation at 400 nm and emission at 505 nm. The initial linear part of the curve was analyzed. For the calculation of active site concentrations, Ac-DEVD-AFC was used for caspases-3 and -7 according to the same protocol, and different amounts of zVAD-fmk were added to all reactions. The ratio between reaction velocity with inhibitor (Vi) over reaction velocity without inhibitor (V0) was plotted against the inhibitor concentration, and the concentration for y = 0 was determined as the active site concentration of the caspase. The specific activity of the caspase enzymes used was determined with a standard curve using free AFC and the Ac-VEID-AFC or Ac-DEVD-AFC substrate for caspase-6 and caspases -3 and -7, respectively. The resulting specific activities were: 11.7 nmol Ac-VEID-AFC/min/mmol caspase-6, 13.7 nmol Ac-DEVD-AFC/min/mmol caspase-3 and 9.9 nmol Ac-DEVD-AFC/min/mmol caspase-7. For the assessment of VEID cleavage in cell lysates, lysates corresponding to 100 mg protein were mixed with an equal volume 200 mM Ac-VEID-AFC in 26 caspase cleavage buffer. Fluorescence was measured as described above. Different concentrations of Ac-VEID-Afc substrate (1-150 mM) were digested with 10 nM caspase-6 in caspase cleavage buffer for 1 h in a black 96well plate (Nunc). Fluorescence was measured every 5 min with excitation at 400 nm and emission at 505 nm. The initial linear part of the curve was analyzed. Different concentrations of lamin A protein (3-400 nM) were digested with 20 nM caspase-6 in caspase cleavage buffer for 15, 30, 45, and 60 minutes at 37uC. The reactions were stopped by shock-freezing, and thawed samples were analysed with the Mesoscale ELISA system. The cleavage rate was calculated at each lamin concentration and plotted against the concentration of Lamin A protein to calculate the Km and Kcat (Fig. S2 ). Caspase-6 activity assays using the Mesoscale ELISA system 100 ng pure lamin A protein was subjected to digestion by different concentrations of caspase enzymes as indicated in the figures (0.05-1000 nM) in caspase cleavage buffer for 30 min at 37uC. The reactions were stopped by shock-freezing, and thawed samples were analysed with the Mesoscale ELISA system: 5 ml of each reaction, corresponding to 25 ng lamin A protein, were spotted onto Mesoscale ELISA plates. Samples were incubated at room temperature for 1 h, then 150 ml 5% BSA in PBS was added per well and the plate was again incubated for 1 h at room temperature. All wells were briefly washed 3 times with 150 ml PBS containing 0.05% Tween-20, and 25 ml antibody mix was added per well (cleaved lamin A antibody, Cell Signaling 2036, 1:100, goat anti-mouse sulfo-tag secondary antibody, MSD technology, 1:500 in 1% BSA/PBS). After 1 h incubation at room temperature, all wells were briefly washed 3 times with 150 ml PBS containing 0.05% Tween-20, and 150 ml 26 reading reagent (MSD technology) was added per well. The plates were then read on a Mesoscale platform electrochemiluminescence reader (MSD technology) according to manufacturer's instructions. For assays using MEF or MCF-7 cell lysates, lysates were diluted in PBS to 0.2 mg/ml, 5 ml were spotted in each well of a Mesoscale ELISA plate and the plate was developed as described above. For assays using neuronal or transfected COS7 cell samples, lysates corresponding to 20 mg total protein were mixed with 100 ng lamin A protein in 16 caspase cleavage buffer. Samples were incubated for 3 h at 37uC, 5 ml were spotted in each well of a Mesoscale ELISA plate and the plate was developed as described above. Mouse embryonic fibroblasts (MEFs) from a C6ko mouse and its C6wt littermate were generated from day 12.5 embryos resulting from timed-pregnant heterozygous breedings and tissues not used for culture were genotyped. The generation and characterization of C6ko mice is described elsewhere [27] . Single pups were dissected in ice cold PBS, the body without head, limbs and liver, lung and heart was minced, the centrifuged pellet was digested with 0.25% Trypsin-EDTA at 37uC for 15 min and neutralized with MEF medium (Dulbecco's modified Eagle medium with high glucose, 10% fetal calf serum, 2 mM L-Glutamine, 100 mM non-essential amino acids, 1 mM sodium pyruvate, 1 mM b-mercaptoethanol). Cells were passaged through a pipette tip, centrifuged and suspended in MEF medium containing DNase I. After centrifugation, cells were suspended in fresh medium, incubated for 2 min at room temperature to let debris and cell clumps settle and cells in the supernatant were seeded into cell culture flasks. At passage 2, cells were immortalized by transfection with pSV3-neo SV40 large T antigen (ATCC) with the Fugene reagent (Roche) according to manufacturer's instructions. Immortalized cells were selected and propagated through the addition of 600 mg/ml G418 to the medium. For the activation of caspase-6, cells were stressed with 50 nM staurosporine for 0-6 h and harvested by trypsinization. Cell pellets were lysed in 50 mM Tris pH 8, 150 mM NaCl and 1% Igepal with 4.2 mM Pefabloc and 'Complete' protease inhibitor cocktail (Roche) on ice, and protein concentrations were determined in the cleared lysates after centrifugation. Cortical neuronal cultures were prepared as described previously [38] from E16.5 littermate embryos obtained from timedpregnant heterozygous breedings and tissues not used for culture were genotyped. Cultures were maintained at 37uC under 5% CO 2 and half of the culture media was exchanged every 4-5 days. At day 10 in vitro, 5 mM camptothecin was added to the media to induce caspase activation [39] , and cells were harvested after 30 h of treatment by scraping in ice-cold PBS supplemented with protease inhibitors (4.2 mM Pefabloc and 'Complete' protease inhibitor cocktail (Roche)). Cells were pelleted by centrifugation and stored at -80uC until lysis. Cell pellets were lysed in 50 mM HEPES pH 7.4, 100 mM NaCl, 1% Igepal, 1 mM EDTA and 10% glycerol with 4.2 mM Pefabloc and 'Complete' protease inhibitor cocktail (Roche) on ice, and protein concentrations were determined in the cleared lysates after centrifugation. Culture, transfection and lysis of COS7 cells COS7 cells were cultured in Dulbecco's modified Eagle medium supplemented with 10% fetal calf serum and 2 mM L-Glutamine. Cells were transfected with human caspase-6 cDNA with a C-terminal DDK tag using the Fugene reagent (Roche) according to manufacturer's instructions. Cells were harvested 0-24 h after transfection by trypsinization. Cell pellets were lysed in 50 mM HEPES pH 7.4, 100 mM NaCl, 1% Igepal, 1 mM EDTA and 10% glycerol with 4.2 mM Pefabloc and 'Complete' protease inhibitor cocktail (Roche) on ice, and protein concentrations were determined in the cleared lysates after centrifugation. Culture and lysis of MCF-7 cells MCF-7 cells were cultured in Dulbecco's modified Eagle medium supplemented with 10% fetal calf serum and 2 mM L-Glutamine. Cells were stressed with 5 uM camptothecin for different amounts of time and harvested by trypsinization. Cell pellets were lysed in 50 mM HEPES pH 7.4, 100 mM NaCl, 1% Igepal, 1 mM EDTA and 10% glycerol with 4.2 mM Pefabloc and 'Complete' protease inhibitor cocktail (Roche) on ice, and protein concentrations were determined in the cleared lysates after centrifugation. Western blotting 20 ng of pure lamin A protein was subjected to digestion by different concentrations of caspase-3, -6, and -7 (0.75-6 mM) in caspase cleavage buffer for 30 min at 37uC. The reactions were stopped by shock-freezing, and thawed samples were run on 4-12% Bis-Tris gels (Nupage, Invitrogen), transferred to PVDF membranes by electroblotting and membranes were developed with primary antibodies in 5% BSA/PBS. Fluorescently labelled secondary antibodies and the LiCor Odyssey Infrared Imaging system were used for detection. For Western blotting of cell lysates, 50 mg total protein were run on 4-12% Bis-Tris gels (Nupage, Invitrogen) and processed as described above. C6wt and C6ko MEFs were seeded at 2500 cells per well into 96well plates. N = 3 for each treatment. The following day, cells were treated with camptothecin (5 mM final) for 16 hours. Post stress, cells were washed in PBS, fixed in 4% formaldehyde (EMS)/PBS for 60 minutes at 4C, washed with PBS, and permeabilized with 0.3% Triton-X 100/PBS. Primary antibody to mouse cleaved lamin A (Cell Signaling Technology) was added at 1:100, in normal goat serum, and incubated at 4C overnight. Plates were washed twice in PBS before adding secondary antibodies, AlexaFluor 488 anti-mouse IgG (Invitrogen), each at 1:800 in normal goat serum, and DNA staining dye, Hoechst 33342 (Invitrogen), at 1:10000, for 90 minutes at room temperature. Plates were washed again twice and left in PBS. Labeled cells were analysed using the ThermoFisher Cellomics ArrayScan VTI, a high content scanning (vHCS) microscope, using version 6.6.2.0 software. The ArrayScan VTI captures images using an ORCA-ER camera in two channels applicable to the assay. XF-53 filters for 405 and 488 nm were used in automated image analysis to quantify nuclei number, and cleaved lamin A, respectively. The Cellomics toolbox, Compartmental Analysis, used an algorithm to encircle perinuclear staining of cleaved lamin A. The primary object in the Hoechst channel was identified as the nucleus of the cell of which 500 total were counted per well. In channels 2, the perinucleus was identified as a ''circ'', which was 3 pixels outside the primary object, where the algorithm measured pixel intensity above background staining. The ''MEAN_AvgCirIntensity'' for each channel was used for analysis, under the vHCS analysis tool, and exported to Microsoft Excel spreadsheets and GraphPad Prism for further analysis. Figure S1 Active site titrations for caspases -3, -6, 7. The exact active site concentration of each caspase used was determined by titrating the enzymes against the pan-caspase inhibitor zVAD-fmk [27] . (TIF) Figure S2 Km determination for lamin A. A: The indicated concentrations of lamin A protein were digested with 20 nM caspase-6 and samples were analysed with the Mesoscale ELISA system. B: Cleavage rates were determined as the slope of the curves in (A) and plotted against the lamin A concentration to obtain values for Km and kcat through curve fitting using the built-in function of the GraphPad Prism 5.0 software package. (TIF)

Quantifying social distancing arising from pandemic influenza

Local epidemic curves during the 1918–1919 influenza pandemic were often characterized by multiple epidemic waves. Identifying the underlying cause(s) of such waves may help manage future pandemics. We investigate the hypothesis that these waves were caused by people avoiding potentially infectious contacts—a behaviour termed ‘social distancing’. We estimate the effective disease reproduction number and from it infer the maximum degree of social distancing that occurred during the course of the multiple-wave epidemic in Sydney, Australia. We estimate that, on average across the city, people reduced their infectious contact rate by as much as 38%, and that this was sufficient to explain the multiple waves of this epidemic. The basic reproduction number, R(0), was estimated to be in the range of 1.6–2.0 with a preferred estimate of 1.8, in line with other recent estimates for the 1918–1919 influenza pandemic. The data are also consistent with a high proportion (more than 90%) of the population being initially susceptible to clinical infection, and the proportion of infections that were asymptomatic (if this occurs) being no higher than approximately 9%. The observed clinical attack rate of 36.6% was substantially lower than the 59% expected based on the estimated value of R(0), implying that approximately 22% of the population were spared from clinical infection. This reduction in the clinical attack rate translates to an estimated 260 per 100 000 lives having been saved, and suggests that social distancing interventions could play a major role in mitigating the public health impact of future influenza pandemics.

Infectious diseases are commonly controlled by minimizing contact between infectious and susceptible individuals. Personal measures to reduce potentially infectious contacts are sometimes referred to as 'social distancing'. It has been suggested that policies encouraging social distancing may be effective against pandemic influenza (Bell et al. 2006; Glass et al. 2006) . It is unclear, however, whether individuals can reduce their infectious contact rate to a level low enough to return a worthwhile public health outcome. An examination of levels of social distancing actually achieved during previous epidemics can provide useful guidance as to the effectiveness of social distancing interventions during future influenza pandemics. The infectiousness of a disease is characterized by the basic reproduction number (R 0 ), which for our purposes is the expected number of infectious contacts per infective when there are no pharmaceutical or behavioural interventions in place and every individual is equally susceptible. More sophisticated definitions are required where individuals have substantially different risks of infection; the methods described by Diekmann & Heesterbeek (2000) are useful in defining and calculating R 0 when contact structures and other kinds of heterogeneity are important. In practice, when an epidemic occurs, the effective reproduction number (R) differs from R 0 due to the deployment of interventions, the build-up of herd immunity and possibly pre-existing immunity. The benefit arising from interventions that additionally decrease R beyond that expected based on herd immunity alone may differ depending on the magnitude of the decrease and its timing (Bootsma & Ferguson 2007; Hatchett et al. 2007 ). If a reduction in the infectious contact rate can be introduced early and sustained, the overall attack rate can be reduced. For a given decrease in the contact rate, the relative reduction in the attack rate is smaller for larger R 0 (figure 1). For example, halving the infectious contact rate may lead to a major epidemic being averted (i.e. a 100% reduction in the attack rate) when R 0 Z2, but, at most, approximately only a 20% reduction in the attack rate if R 0 Z4 (figure 1). It is more realistic to assume that interventions to reduce R cannot be sustained indefinitely. If interventions are introduced, and subsequently removed before herd immunity has increased sufficiently to reduce R to approximately 1, this will postpone and diminish the peak incidence of the epidemic (though not necessarily the eventual attack rate), thus reducing the peak load on health services. Finally, we will argue that if the introduction of timelimited interventions (e.g. social distancing) is timed in such a way as to minimize the number of active infective cases as R approaches unity, then the minimum achievable attack rate can be obtained. Through a combination of geographical isolation and public health measures, the city of Sydney, Australia, delayed the introduction of the Spanish flu by several months until early 1919, at which point public health officials responded almost immediately (McCracken & Curson 2003) . As with many populations affected during the 1918-1919 pandemic (e.g. Geneva, Switzerland; Chowell et al. 2006) , Sydney experienced multiple epidemic waves. There are several theories explaining the multiple waves, including transient post-infection immunity, viral antigenic drift and the involvement of multiple viral strains; substantial counterarguments exist for all these theories and the issue remains unresolved (Taubenberger & Morens 2006) . In the case of Sydney, the beginning of a second wave coincided with the lifting of public infection control measures, suggesting that transient adoption of social distancing measures could underlie the observed dynamics (McCracken & Curson 2003) . More broadly, Hatchett et al. (2007) observed that the quality and timing of non-pharmaceutical public health interventions aimed at decreasing disease transmission by reducing social contact rates appeared to influence the course of influenza epidemics in 17 large US cities during 1918, with second waves occurring only after the relaxation of interventions. We hypothesize that the public of Sydney in 1919 initially responded to the public health measures and subsequently rising and/or high incidence of cases and, particularly, case fatalities by reducing their exposure to potentially infectious contacts. Bootsma & Ferguson (2007) have documented a similar reactive reduction in contact rates in response to high mortality rates arising from pandemic influenza. As the perceived risk decreased, the public subsequently relaxed, returning to normal behaviour. There is a delayed negative feedback between the contact rate and the incidence, and, as with many dynamical systems that experience time lags, oscillations develop. We assume that R 0 is constant over the duration of the epidemic. This is in contrast to Chowell et al. (2006) for example, who assumed that R 0 differed between waves-we consider this to be a phenomenological rather than explanatory assumption. In this paper, we seek to estimate the degree of social distancing that occurred in Sydney in 1919. To do this, we use the epidemic curve and other historical data to estimate (i) the disease reproduction number over the course of the 1919 Sydney influenza epidemic, (ii) bounds on the fraction of people who were asymptomatic seroconverters (whether infectious or not) in that epidemic and (iii) bounds on the fraction of people who were resistant before the epidemic began (e.g. owing to heterotypic immunity). The methods used in this paper are described in three sections. Section 2 establishes the relevant aspects of the historical background, including why we argue for attributing the epidemic waves to the effect of social distancing. Section 3 measures the reproduction number on each day of the Sydney epidemic by applying the method of Wallinga & Teunis (2004) . Section 4 presents methods for using the observed reproduction numbers and the cumulative number of cases to derive relationships between the serological attack rate and the initial fraction of the population that are susceptible. Each of these quantities has direct policy implications for an epidemic. They are often incorporated into models (e.g. Ferguson et al. 2005; Longini et al. 2005) , despite considerable uncertainty about which values are appropriate for pandemic influenza. In this section, we describe the history of the epidemic in Sydney and what is known about the population's behaviour at each stage. The method we subsequently present in §5 relies on using the historical record to identify periods during the epidemic when the population behaved normally with regard to the transmission of disease. We assume that the public's willingness to reduce transmission relies on their perception of the risk associated with the epidemic. We argue that the historical record, as described by McCracken & Curson (2003) , shows periods during which the perceived risk would be high (owing to high infection incidence or the imposition of control measures), and periods when the risk would be perceived as low. Three periods (labelled A, C and E) are associated with a high perceived risk and three others (B, D and F) are associated with a low perceived risk, and consequently normal transmission. Figure 2a shows a summary of these periods and a detailed explanation follows. If the intervention is not introduced immediately and sustained indefinitely, a lower reduction will be achieved. We define period A as beginning from the time when the first cases were identified (27 January 1919). During this period, extensive infection control measures were imposed, including: closing theatres and public places of entertainment; compulsory wearing of masks on all public transport and in public places; closure of schools; prohibition of race meetings and church services; and removal of patients to hospital and strict quarantine of contact (see McCracken & Curson (2003) for a complete list). As the incidence remained low in comparison with severe epidemics reported from elsewhere around the world, authorities deemed that the threat had passed and most measures were lifted on 1 March. From 1 March until the reimposition of control measures on 24 March (period B), the incidence rose exponentially. Even so, the daily death rate was low in absolute terms (figure 2a) because initial incidence was low, and the mean delay between symptom onset and death was 8.5 days (Armstrong 1920) . During this period, we assume that the population approached normal behaviour. Things changed on the weekend of 22-23 March, when 20 people died of influenza; infection control measures were reimposed around the end of March. We assume that, from 25 March, the perceived severity of the disease was high enough to reduce transmission. These measures were continued throughout the first wave (period C). We assume that the decreasing incidence led to a decreased perceived risk and that the public started to resume normal behaviour as the authorities lifted infection control measures in the middle of May. We assume that behaviour approached normal during the period D, 25 March to mid-June. A second wave began shortly after the infection control measures were lifted (i.e. during period D), and was clearly apparent by mid-June. Even though infection control measures were not reimposed, we assume that the high incidence was a sufficient threat to alter people's behaviour. We define this period of altered behaviour (period E) as running from mid-June to 12 August. We assume that people resumed normal behaviour by 12 August (thereafter period F), as by then the incidence of hospitalizations and deaths had dropped substantially, and the number of hospitalizations ceased to be reported in daily papers. We argue that social distancing is an appropriate explanation for the waves for several reasons. Seasonal changes in virus transmissibility, while possible, cannot be of sufficient size to cause multiple waves-particularly over such a short time period. Indeed, seasonal influenza epidemics on an annual basis cannot occur if the difference attributable to seasons is more than approximately 10% of R. Multiple circulating viruses may have contributed to the waves in Europe, where repeat infection was documented (Ministry of Health 1920). However, this could not have occurred in Sydney, where reinfection was extremely rare, and when it did occur the symptoms were mild (Armstrong 1920 ). Armstrong reports that 814 out of 1488 (55%) health care workers were attacked once, yet only four of these (0.5%) were recorded as being attacked twice. It might be argued that the first and second waves in Sydney were caused by different strains which provided cross-protection. For this to produce two comparable waves would require that the second strain be substantially more infectious (higher R 0 ) than the first to overcome the effects of herd immunity. We will show that the reproduction numbers during both waves in Sydney were remarkably similar. Finally, if applied in a transient manner (i.e. applied then lifted too early), there is an underlying mechanistic explanation of resulting waves (Bootsma & Ferguson 2007) . Daily hospital admissions attributable to influenza were collated from the Sydney Morning Herald, which published a daily report except that data for weekends were not broken into separate days. Daily data on deaths attributable to influenza came from the New South Wales Statistical Register 1919-1920 (table 105) . These data have already been given in figure 2. Land and sea border control/quarantine surrounding Sydney meant that the overwhelming majority of cases were not imported. At the height of the epidemic, the Sydney hospitals were overloaded and turned away patients who would have otherwise been admitted (McCracken & Curson 2003) . During the period where the hospitals were not overloaded, the epidemic curve and time-dependent effective reproduction number (see §3.2) can be inferred from either the hospitalization or death data. We estimated the effective reproduction number R(t) for each day of the epidemic using the method of Wallinga & Teunis (2004) . The method assumes that the infectiousness function, which describes the rate at which an infected individual transmits infection over the course of their infection (Becker 1989) , is known. We derived an average infectivity profile from Ferguson et al. (2005) and defined b(a) to be the average relative infectivity of a person on day a of their infection; transmission was assumed not to occur after 10 days. The mean serial interval arising from the resulting infectivity profile was 2.6 days. The method was applied separately to both the death and hospitalization data. Strictly speaking, the method of Wallinga & Teunis (2004) should be applied to incident infections. As infectious events are rarely observed, we (and previous authors) must use symptom onset, death or some other measure as a surrogate marker for infection. Two issues arise, which are as follows. First, notifications of markers (e.g. deaths) may be substantially thinned versions of incident cases. Second, there is a delay, most likely of variable duration, between the infection and the chosen marker. Wallinga & Teunis (2004) showed that a small degree of thinning (e.g. resulting from under-reporting of cases) would not bias estimates of R(t), but did not investigate the effect of using only a small fraction of cases (as when using deaths as a surrogate when the case-fatality rate is low) to estimate R(t). In the Sydney 1919 epidemic, the probability of hospitalization and death for a given clinical infection was 4.8 and 1.2%, respectively. We used repeated stochastic simulations of an epidemic with R 0 in the range of 1.5-2.5 in a population of 800 000, with the number of daily cases thinned to 5 and 1% to confirm that thinning per se results in no discernible bias in the resulting estimates of R over the course of an epidemic. If the delay from the infection to the chosen marker (e.g. death) is fixed, there is no bias in the resulting estimates of R(t). Conversely, if there is variability in the delay, then there is a potential for bias, particularly if the distribution of the delay is right-skewed. The effect of the time-to-marker delay distribution is to widen the epidemic curve of the marker, relative to the true incidence curve. The wider the distribution from the infection time to the marker, the greater the potential bias. On theoretical grounds, it is easy to show that during the early and late exponential phases of an epidemic (i.e. its leading or trailing edge), every marker gives an unbiased estimate of R, provided that the exponential phase is itself long in duration compared with the width of the distribution for the marked event. During the peaks of the epidemic, the epidemic curve is not exponential and the above result does not apply. In our case, Armstrong (1920) provided data on the distribution of time from the onset of influenza symptoms to death (mean 8.86 days, s.d. 6.0 days), which is well described by a gamma (kZ2.74, qZ3.23) distribution. We repeated our epidemic simulations; this time, modelling the time from infection to death using this gamma distribution shifted 1.5 days to the right to account for the disease incubation period (assumed fixed). Applying the method of Wallinga & Teunis (2004) to the death data confirmed that the resulting daily estimate of R has little discernible bias during the early exponential growth period and again during the final days of the epidemic. Our application of the method to death data does underestimate R during the middle of the epidemic, and overestimate it at the start of the declining phase; the extent of the bias increases with increasing R 0 , though it is less than 10% for a freely evolving epidemic with R 0 Z1.5. During periods when R is close to 1, the bias is also small-this is reflected in the similarity of the hospitalization and death results. Given that we are predominantly interested in the reproduction numbers during the periods of early exponential growth and during the final cases of the epidemic ( §4.5), we consider that the method of Wallinga & Teunis (2004) produces estimates of R that are adequate for our purposes. To remove dayto-day variation in estimates of R(t) for the purpose of making inference, we fitted a smooth curve to the daily estimates of R(t) using cubic splines with knots every 7 days. Figure 2b shows the daily estimates of the effective reproduction number R(t) based on both the hospitalization and death data. The estimates are noisier during the periods when case numbers are small (e.g. before day 75 and after day 200). As expected,RðtÞ begins above 1 and drops below 1 as the first wave peaks, though not by much ðR min ðC ÞZ 0:85G0:01ðGs:e:ÞÞ. It returns to greater than 1 at approximately day 130 (figure 2b), which is approximately when the second wave of the epidemic began to grow, and remained above 1 until day 165, the peak of the second wave. It is apparent that RðtÞ based on the hospital admissions underestimates R(t) during both waves due to hospitals being overloaded (figure 2b). At times other than early in the epidemic when the number of deaths is very small, the estimates of R(t) based on either hospitalizations or deaths are very similar (allowing for deaths to lag hospitalized cases; figure 2b). We henceforth use deaths only to make inference on R(t). Indeed, we expect there to be less bias in the estimates of R(t) arising from deaths compared with hospitalizations. This is because being admitted to hospital is dependent on many factors unrelated to the epidemiology of disease that may vary over time (e.g. perceived need for hospital care based on the case-fatality rate). The maximum value of the smoothed curve during period B gave an estimate ofRðBÞZ 1:59G0:02ðGs:e:Þ (figure 2b). The mean of the daily reproduction number in period F waŝ RðFÞZ 0:95G0:04ðGs:e:Þ. In this section, we present a method for inferring the degree of social distancing during different periods of the epidemic. Our method relies on knowing the reproduction number operating at each time (established in §3). We attribute part of the variation in this reproduction number to herd immunity and the remainder to social distancing. The total population size of Sydney was NZ810 700, of which at least 14 130 (1.74%) were admitted to hospital and approximately 3500 (0.43%) died as a result of influenza infection (McCracken & Curson 2003) . Based on a survey of 600 establishments covering 106 923 employees, the proportion of workers that were absent from duty as a result of influenza was 36.6% (Armstrong 1920, p. 144 ). This was considered as an unbiased estimate of the clinical attack rate, although we argue that the serological attack rate (proportion of workers who developed resistance) may have differed. We denote the proportion of the population that were recorded as being hospitalized or as having died on day t as h(t) and d(t), respectively; these are known from the data. We denote the proportion susceptible as s(t), and the per capita incidence on day t as i(t). We do not assume that infectives were necessarily symptomatic, but they are all assumed to have become immune. Our model assumes that mixing within the population can be approximated as homogeneous. We assume a form for the effective reproduction number that incorporates the build-up of immunity in the population and social distancing, where s(t) is a scalar, which describes the extent to which behaviours resulting in disease transmission are maintained. A reduction in s(t) indicates that disease transmission has decreased for some reason other than the depletion of susceptibles. For example, when the population is behaving normally (i.e. no social distancing), s(t)Z1, and when potentially infectious contacts are reduced by half, s(t)Z0.5. We consider that the population closely approached normal behaviour during periods B and F, and possibly during period D, i.e. s(B)Zs(D)Zs(F)Z1 (table 1) . Our aim is to use this model to estimate s(t) by estimating R(t) and s(t). More specifically, we seek to estimate R A ðtÞ Z R 0 sðtÞ Z RðtÞ sðtÞ ; ð4:2Þ which we refer to as the 'adjusted reproduction number'the adjustment referring to the correction of the effective reproduction number for the proportion of the population that are susceptible. When there is no social distancing, R A ðtÞZ R 0 . Our goal is to estimate how much of the variation in the reproduction number exceeds that which can be attributed to the build-up of immunity, and to attribute that to social distancing. We define s min to be the lowest value of s(t) obtained from the analysis, corresponding to the point of greatest social distancing. The serological attack rate (final proportion infected and developing solid immunity) is aZs(0)Ks(N). The fraction of the population remaining susceptible at time t is equal to the initial proportion susceptibleKthe cumulative proportion infected by t, sðtÞ Z sð0ÞK ð t 0 iðt 0 Þdt 0 : ð4:3Þ We do not observe i(t) and must infer it from the daily death and/or hospitalization data. In the case of deaths (which in §3.3 we show yields the best estimate of R(t)), we must account for the time delay (t) between infection and death. The time from symptom onset to death was remarkably similar across all age groups with a mode of 7 days (Armstrong 1920; figure 3 ). We add 1.5 days for the incubation period (Ferguson et al. 2005 ) and round to the nearest integer, so that tZ9. Hence, re-expressing equation ( are compatible with the observed reproduction number over the course of the epidemic. In this section, we discuss the possible range of values of the serological attack rate. During the preparation of this paper, a similar theory has been presented (Bootsma & Ferguson 2007 ), which we present in more detail. If social distancing is sufficiently effective (s!1/R 0 ) and can be maintained, then an epidemic will go extinct by the epidemic threshold theorem (Becker 1989) . In a large population, the fraction who become infected in this case is negligible. This may have contributed to the extinction of SARS virus (Riley et al. 2003) . If an epidemic cannot be contained by social distancing, and goes on to infect a sizeable fraction, the serological attack rate a must lie between a minimum value a min and a maximum value a max . Consider two hypothetical major epidemics, the first without social distancing and the second with what we will argue is optimum effective social distancing. For an epidemic in a reasonably well-mixed population, unimpeded by social distancing, a max is obtained from R 0 and s(0) by the final size equation (Diekmann & Heesterbeek 2000) a max Z sð0Þð1Ke Ka max R 0 Þ: ð4:5Þ Given estimates of R 0 and s(0), we use equation (4.5) to obtain an estimate of this maximum serological attack rate ðâ max Þ, noting that this estimate is quite robust to a range of underlying spatial contact structures and variation in infective potential among individuals (Ma & Earn 2006) . In the second scenario, we assume that the eventual extinction of the epidemic is a result of the development of resistance in the wider community. The optimum attack rate is obtained by applying social distancing such that as the proportion of susceptibles in the community falls below 1/R 0 , the number of infectives is so small that the epidemic fades out without infecting a significant fraction of the remaining susceptible population. Here, the ultimate proportion of the population remaining susceptible is s(N)Z1/R 0 (if s(N)O1/R 0 , reintroduction of the infection could lead to another epidemic wave). This condition allows us to define a min Zs(0)K1/R 0 and, given estimates of R 0 and s(0), we can obtain an estimate of this minimum attack rate ðâ min Þ. One would think that achieving this limit in practice should be rather difficult due to its extreme nature. The difference between the scenarios arises due to the following reasons. Once the proportion of susceptibles in the community falls below 1/R 0 , the effective reproduction number drops below unity regardless of the degree of social distancing, and the epidemic is doomed to extinction. A freely flowing epidemic, however, overshoots a min because at this stage the largest number of infectives is active. In the optimal case, social distancing is used to minimize the number of infectives at this stage, so that there is no overshoot. The ultimate attack rate therefore depends on how many individuals are infected as R(t) crosses 1. In Sydney 1919, the attack rate must have lain between these extremes: a min % a% a max . Based on the difference between our estimates of a and a max , we scale up the number of lives actually lost to estimate how many lives might have been lost if the epidemic had been entirely unimpeded by social distancing. By this measure, the number of deaths per 100 000 of the population that were prevented by social distancing (D) was DZ ða max =a K1Þp !10 5 . Whether or not social distancing has occurred during an epidemic, if it is relaxed (i.e. s(N)Z1) during the final cases, it follows from equation (4.1) that RðNÞ Z R 0 ð1KaÞ: ð4:6Þ Under optimal social distancing with a minimum possible attack rate ðaZ sð0Þ K 1=R 0 Þ, we expect R(N) to be unity. In epidemics where transmission is unimpeded (s(t)Z1 throughout), epidemic decline is much more rapid. During the final phase, there are sufficient infectious cases in that many susceptibles are infected, even though the reproduction number is well below 1. To estimate the reproduction number during periods B and D (RðBÞ andRðDÞ, respectively), we took the maximum of the smoothed estimate of R(t). Our estimate of the final reproduction number (RðFÞ) was the mean of the daily estimates during period F, weighted by the number of deaths on that day. By substituting equation (4.4) into equation (4.1) and solving for s(0) after setting s(B)Zs(F)Z1, we obtain a relationship between s(0) and a along with the reproduction numbers during periods B and F and the associated cumulative number of per capita deaths, Here, t B refers to the time until the peak in the effective reproduction number during period B, and t F is the time to the middle of period F. We could have additionally usedRðDÞ; however, a priori we were less confident that the population was behaving normally during period D. All analyses were undertaken using the computing environment R v. 2.5.0 (R Development Core Team 2007). Having estimated the values ofRðBÞ andRðFÞ, equation (4.7) establishes a unique relationship between a and s(0). The reported clinical attack rate is the obvious first choice as an estimate of a, but may be biased for several possible reasons: (i) it is conceivable that clinical cases may not have conferred solid immunity, (ii) cases that seroconvert may be asymptomatic, and (iii) illness may have been mistakenly attributed to influenza when it was in fact caused by another influenza-like illness (e.g. respiratory syncytial virus). Hence, we explore the values of a to be an arbitrary 10% below (0.329) and 10% above (0.403) the reported clinical attack rate. The upper value turns out to be just above the maximum possible under our final approach to estimating a; that is, to use equation (4.7) under the assumption that everyone was initially susceptible to infection (i.e. s(0)Z1). We therefore explore three estimates of a. For each estimate, we compute the corresponding values of s(0), a max , a min , R A (B), R A (D), R A (F) and s min . We estimate R 0 as the average of R A (B), R A (D) and R A (F). Setting the serological attack rate to the observed clinical attack rate of 0.366 estimates the initial susceptible proportion to be s(0)Z0.912 andR 0 Z 1:76 (table 2) . Setting the serological attack rate to aZ32.9% (i.e. 10% lower than the clinical attack rate) corresponds to an initial susceptible proportion of s(0)Z0.821. This scenario requires that 10% of those who developed clinical symptoms were not solidly protected against future severe attack, contradicting contemporary observations of influenza-dedicated hospital staff (Armstrong 1920) . If the population was initially fully susceptible (s(0)Z1.0), a serological attack rate of aZ0.401 is required to explain the epidemic dynamics. Again, if we assume that 0.366 is an accurate measure of the clinical attack rate, then it follows that 8.7% of those infected developed immunity without having developed clinical symptoms to the extent that they did not attend work. Although we do not give credence to a scenario that assumes s(0)Z1.0, as it is probable that there was at least some heterotypic immunity from seasonal influenza, we note that it creates an upper bound of 8.7% for the fraction of infectives who could have been asymptomatic transmitters. We suggest that, of these three estimates, the survey-based estimate of the clinical attack rate (0.366) is probably closest to the true value of the serological attack rate (i.e. aZ0.366) and hence our preferred estimate of R 0 is 1.76 (table 2; figure 4) . Each of the three scenarios returns the same value of s min (this is a mathematical consequence of our methods), corresponding to a reduction in the infectious contact rate of 38% during the first wave (table 2) . During the second wave, the maximum estimated reduction in the infectious contact rate was less (24%). Interestingly, the second wave was perceived as being more severe than the first, so the difference between these values may be attributable to the public health policy of encouraging social distancing during the first wave. Alternatively, the difference could be explained by the exceptionally heavy rain that fell nearly throughout the month of May (following the first wave), thus discouraging people from getting out and circulating in the wider population (McCracken & Curson 2003) . Assuming homogeneous mixing, no social distancing, s(0)Z31.2% and R 0 Z1.76, using equation (4.5), we would expect an attack rate of 58.8%-much greater than the 36.6% observed. Assuming that the number of deaths is directly proportional to the attack rate, the reduction indicates that DZ260 per 100 000 lives were possibly saved as a result of social distancing. The estimated value of a min was approximately 6% less than the modelled serological attack rate for the three parameter combinations examined. This suggests that few additional lives could have been saved by increasing the degree of social distancing, unless it was able to eliminate the epidemic. The observation that R(t) reduces to near 1 for a prolonged period during the last days of the epidemic further supports the conclusion that a was close to a min . Substituting aZ0.588 into equation (4.6), the expected reproduction number during the final stages of the epidemic is 0.725substantially less than the 0.95 observed. Table 2 . Values of the attack rate a and the corresponding values of the initial susceptible proportion (s(0)), the basic reproduction number (R 0 ), the minimum and maximum fractions that could have been infected (a min , a max ), the adjusted reproduction numbers during periods when we expect that social distancing is at a minimum (R A (B), R A (D) and R A (F)), the social distancing coefficient when social distancing was at its greatest (s min ), and the estimated number of deaths avoided per 100 000 (D). (Values in the first row are computed by assuming that a was 10% less than the reported clinical attack rate with s(0) allowed to vary freely. The second row is computed using the clinical attack rate as an estimator for a. The third row is computed by adjusting a to obtain s(0)Z1.0.) The relationship between the adjusted reproduction number and the number of daily deaths for the first and second waves shows a negative trend-more deaths mean greater social distancing (figure 5a,b). For figure 5c,d, we wish to plot the reproduction number against the number of infections on the same day. Since the number of infections is unknown, we use the number of deaths 9 days later as a proxy. The clockwise cycles reveal the delay between the infection and the subsequent decline in R A (t)-and hence the degree of social distancing. We have assumed an 'all or nothing' model of prior immunity, meaning that a fraction of individuals were totally protected from infection during the pandemic period. The main alternative model of prior immunity is that a fraction of the population is partially immune, having a lower (but non-zero) risk of infection. Under some circumstances, there will be material differences between the behaviour of these prior immunity models: if R 0 is very large, all susceptibles, whether fully or partially immune, will inevitably be infected; alternatively, if there is assortative mixing between classes of susceptibles, fully susceptibles will be overrepresented during the early stages of the epidemic and underrepresented in later stages. These circumstances do not apply to the Sydney 1919 epidemic-there was a reasonably low attack rate (less than 50%) and little evidence to support strongly assortative mixing. While our model result is that 10% of the population were fully immune, for these data we cannot easily distinguish this from alternatives, such as where 20% of the population had 50% of the normal risk of infection. While the infectivity profile we use has empirical support, it is interesting to consider the effect of changing the infectivity profile. Had we used an infectivity profile with a shorter mean serial interval, we would have obtained reproduction numbers closer to 1, meaning that smaller changes in the degree of social distancing would explain the epidemic waves. However, the reproduction number cannot be reduced much below 1.6 before it becomes impossible to achieve an attack rate of 36.6%, in an epidemic with two waves of similar magnitude. On the other hand, a longer serial interval would have produced higher estimates of R 0 . In this case, we have underestimated the social distancing achieved during the 1919 epidemic. It is possible that other interventions, such as closing schools and quarantining infectives, played a role in containing the epidemic. We argue that most of these can be broadly categorized as social distancing. Measures such as quarantine are likely to have been practised more or less constantly throughout the epidemic and probably did not contribute to the changes in R(t). We conclude that the variable application of social distancing, whereby individuals reduced their infectious contact rate in response to the perceived risk, is a plausible explanation for the multiple waves of pandemic strain influenza seen during 1919 in Sydney, Australia. Indeed, while the waxing and waning of the multiple waves appears dramatic, the degree of social distancing required to explain this (in this case, at most, halving one's infectious contact rate) seems quite possible. More generally, Bootsma & Ferguson (2007) and Hatchett et al. (2007) have demonstrated that variation in the timing of introduction and lifting of non-pharmaceutical interventions aimed at reducing contact rates can explain why cities experienced different inter-wave periods, ranging from being so short as to be undetectable through to several months (Taubenberger & Morens 2006) . We note, however, that transient social distancing certainly does not explain why the case-fatality rate of the 1918-1919 pandemic typically was higher during the second wave, as indeed was the case for Sydney (McCracken & Curson 2003) . However, note that the very similar reproduction numbers observed during both waves of the epidemic support our initial assumption that R 0 did not differ over the course of the epidemic. Subject to the assumption that infection at any time conferred protection against a subsequent severe attack, we conclude that approximately 9% of the population were resistant to the epidemic strain prior to the epidemic, and that, during the epidemic, not more than approximately 9% of infections that conferred resistance to the epidemic strain were subclinical to the extent that people were able to continue working. Using our best estimate that 91.2% of individuals were initially susceptible, the R 0 of the 1919 influenza epidemic in Sydney was 1.8, consistent with recent estimates that have used a similar mean serial interval (Ferguson et al. 2005; Sertsou et al. 2006) . The observed attack rate, however, was substantially less than would be expected for this basic reproduction number, and we argue that social distancing is a plausible reason for this. This result underlines the effective role that social distancing could possibly play in mitigating the effects of a future pandemic of influenza.

Novel Inhibitor Design for Hemagglutinin against H1N1 Influenza Virus by Core Hopping Method

The worldwide spread of H1N1 avian influenza and the increasing reports about its resistance to the current drugs have made a high priority for developing new anti-influenza drugs. Owing to its unique function in assisting viruses to bind the cellular surface, a key step for them to subsequently penetrate into the infected cell, hemagglutinin (HA) has become one of the main targets for drug design against influenza virus. To develop potent HA inhibitors, the ZINC fragment database was searched for finding the optimal compound with the core hopping technique. As a result, the Neo6 compound was obtained. It has been shown through the subsequent molecular docking studies and molecular dynamic simulations that Neo6 not only assumes more favorable conformation at the binding pocket of HA but also has stronger binding interaction with its receptor. Accordingly, Neo6 may become a promising candidate for developing new and more powerful drugs for treating influenza. Or at the very least, the findings reported here may provide useful insights to stimulate new strategy in this area.

In recent years, severe flu-like human cases were reported around the world and subsequently the causative virus was identified as the influenza A virus [1, 2] . The virus was spreading rapidly around the world and had been identified as a new reassortant with three genetic lineages, mainly with a swine origin. Therefore, it was called swine-origin influenza virus (S-OIV). Owing to its extremely rapid human-to-human transmission rate, within only two months the 2009 S-OIV had been detected throughout the entire world. On June 11th, 2009 the World Health Organization (WHO) declared an official pandemic, the first pandemic in the 21st century [3] . Influenza A virus that belongs to the Orthomyxoviridae family is a negative-strand segmented RNA virus, in which the surface membrane proteins are constituted by three important components: M2 proton channel, hemagglutinin (HA), and neuraminidase (NA). The M2 proton channel is responsible for proton conductance vitally important to viral replication. HA is responsible for binding to the surface of the infected cell as a trimer leading to the attachment and subsequent penetration by viruses into the target cell. NA is responsible for cleaving the terminal sialic acid moieties from the receptors to facilitate the elution of the progeny virions from the infected cell [4] . Therefore, any of the three components can be the target for drug design against influenza virus. Recently, stimulated by the successful determination of its high-resolution three-dimensional structure [5] , many discussions about the M2 channel have been made in this regard [5, 6, 7, 8, 9] . The two existing M2 drugs, amantadine (Symmetrel) [10] and rimantadine (Flumadine) [10] approved by FDA, are no longer effective because of their inefficacies to influenza virus. Sialic acid (SA) as a natural ligand combines with both of the glycoproteins (HA and NA) and located at the membrane of host cell, which is the basis of heme-agglutination when viruses are mixed with blood cells and entry of the virus into cells of the upper respiratory tract [11, 12] . According to the mutagenic analysis the residues of both HA1 and NA binding sites are quite conserved for most influenza A strains [13, 14] . Owing to its deep active site cleft, the NA has been an attractive target for drug design. Both zanamivir and oseltamivir were designed by modifying the sialic acid (SA) structure. The two FDA-approved clinical drugs were once successfully used to inhibit the spread of influenza viral progeny [15] by binding to viral surface glycoprotein of neuraminidase (NA) [15] . However, it has also been found from several clinical cases [16, 17, 18] that oseltamivir failed to treat avian influenza virus. It is both antigenic drift (sequence base mutations) and antigenic shift (genetic recombination) of segmented RNA genome of influenza viruses that have caused the NA inhibitor being resistant [19, 20] . HA facilitates viral entry through binding to the host surface sialic acid residues [21] . Accordingly, if HA is blocked at its sialic acid binding site by a small molecule, the viral entry process will be stopped and the penetration of viruses into host cell prevented. In comparison with NA inhibitors, the HA inhibitors were usually more effective in inhibiting influenza virus. For all the HA subtypes (H1-H16) so far identified [22] , the HA1 subtype from the recent pandemic H1N1/09 virus was taken as the target for constituent screening and drug design [23] . Despite of many year scientific research efforts, so far there is no clinical available inhibitor against HA1. On the other hand, many studies have indicated that computational approaches, such as structural bioinformatics [24, 25] , molecular docking [26, 27] , pharmacophore modeling [28] , identification of proteases and their types [29] , and HIV protease cleavage site prediction [30, 31] , can timely provide very useful information and insights for drug development. Encouraged by the aforementioned studies, the present study was initiated in an attempt to find a new antiinfluenza compound by screening the fragment database for the optimal constituent inhibitor. Meanwhile, the techniques of the core hopping with glide docking and molecular dynamic simulation were also utilized to analyze the binding interactions between the inhibitor and HA1, in hopes that the findings thus obtained will be useful for developing new and powerful drugs against H1N1 influenza virus. The crystal structure for the HA1 subtype from the recent pandemic H1N1/09 virus was downloaded from the PDB Bank [32] . Its PDB ID is 3AL4. The antigenicity of the HA1 from the swine-origin A (H1N1)-2009 influenza A virus is quite similar to that of the HA from the 1918 pandemic virus [23] . The bindingsite was identified by the SiteMap tool in Schrodinger Suite 2009 (www.schrodinger.com) as described in [34, 35, 36] . The bind-site encompassed the ligand N-Acetyl-D-Glucosamine (NAG), which is observed in complex with HA1 of the crystal structure (PDB: 3AL4). Shown in Fig. 1 is a close-up view for the binding site of protein HA1 rendered by the molecular surface. The binding pocket is formed by those residues that have at least one heavy atom (i.e., an atom other than hydrogen) with a distance ƒ5 Å away from any heavy atom of NAG ligand when it is bound to the receptor at the binding site, as elaborated in [37] . The segments of loop1, loop2, loop3, and loop4, which play an important role in the interactions with the ligand, are shown by four different colors with their respective key residues: Ser92, Glu72, Pro143, and Arg227 ( Fig. 1) . The motions of such four residues were monitored during the molecular dynamics simulations. The drug-like database and the fragment database derived from ZINC [38] were used for virtual screening and core hopping searching, respectively. The Glide5 docking program [39] interfaced with Schrodinger Suite 2009 [33] was used to screen the drug-like database from ZINC [38] based on the 3D structure of 3AL4. The preparation and refinement protocols for protein receptor and all compound structures were performed on the Protein Preparation Wizard and LigPrep modules embedded in Schrodinger 2009 [33] , respectively. For protein preparation, the process included assigning bond orders, adding hydrogen, treating metals, treating disulfides, deleting waters and alleviating potential steric clashes, adjusting bond order and formal charges by protein minimization with the OPLS2005 force field [40] , the constrained refinement value of RMSD for the protein was limited to 0.3 Å . Meanwhile, for the compounds, the preparation consisted of the generating possible states by ionization at target pH 7.062.0, desalting, retaining chiralities from 3D structure and geometry minimization with the OPLS2005 force field [40] . When the above steps were accomplished, all investigated compounds were docked into the receptor pocket through the rigid docking model with the Standprecision (SP) scoring function [41, 42] to estimate the binding affinities. Many useful clues for drug design can be achieved through molecular docking studies (see, e.g., [24, 27, 43, 44, 45, 46] ). In order to gain even more useful information in this regard, the novel drug design algorithm called ''Core Hopping'' [33] was used in this study that has the function to perform both the fragment-based replacing and molecular docking. Such method is particularly useful for de novel drug design because it can improve the activity of the template, which was ZINC01602230 compound in this study. As a lead compound screened out from the drug-like database, the template was taken as an initial structure to subject to optimization via the core hopping method by finding the optimal cores that are attached to the scaffold part of the template in binding with the protein receptor. During the process of core hopping, the first step was to define the points at which the cores were attached to the scaffold. It was performed in the Define Combinations Step from the Combinatorial Screening panel [33] . The second step was to define receptor grid file, which was done in the Receptor Preparation panel [33] . The third step was to prepare the cores attached to the scaffold for the fragment database derived from ZINC [38] . Finally, the cores thus obtained were sorted and filtered by goodness of alignment and then re-docked into the receptor after attaching the scaffold, followed by using the docking scores to sort the final molecules. Many marvelous biological functions in proteins and DNA and their profound dynamic mechanisms, such as switch between active and inactive states [47, 48] , cooperative effects [49] , allosteric transition [50, 51] , intercalation of drugs into DNA [52] , and assembly of microtubules [53] , can be revealed by . The binding pocket is defined by those residues that have at least one heavy atom with a distance 5 Å from the NAG ligand [37] . The four loops (loop1, loop2, loop3, and loop4) that play an important role in interacting with the ligand are represented by round ribbons of four different colors as well as their key residues Ser92, Glu72, Pro143, and Arg227, respectively. The motions of such four residues were monitored during the molecular dynamic simulation. The docked poses for ZINC01602230, Neo and Neo6 are shown with the stick model colored in purple, yellow and dark green, respectively. doi:10.1371/journal.pone.0028111.g001 studying their internal motions [54] . Likewise, to really understand the action mechanism of a receptor with its ligand, we should consider not only the static structures concerned but also the dynamical information obtained by simulating their internal motions or dynamic process. In order to examine whether the designed inhibitor remains bound in the presence of explicit solvent from a dynamic point of view, the molecular dynamic simulation was performed with GROMACS 96-53a6 force fields [55] with the periodic boundary conditions (PBC) by using GROMACS 4.0 package for Linux. The topology files and charges for the ligand atoms were generated by the Dundee PRODRG2.5 Server (beta) [56] . Before starting the simulations, all the models were solvated with the explicit simple point charge (SPC) water in a cubic box. The models were covered with a water shell of 1.0 nm from the surface of the protein. The system was neutralized with six chlorine ions to replace the six SPC water molecules. Subsequently, the energy minimization was performed for the system concerned by using the steepest descent until touching a tolerance of 100kJ/mol. And then, the 10 ns MD simulations were carried out with a time step of 1 fs; the corresponding coordinates were stored every 100 fs. The PME algorithm was used to calculate the electrostatic interactions. All simulations were run under the periodic boundary condition with NVT ensemble by using Berensen's coupling algorithm for keeping the temperature at 310 K and pressure at 1atm. All bonds were constrained by using the LINCS algorithm. The GROMACS 4.0 package was utilized to analyze the results. The drug-like database from ZINC [38] was screened by using Glide5 for its near-optimal performance aimed on targeting the HA1 receptor (PDB ID:3AL4). The top hit (ZINC01602230) or (2amino-N-(7H-purin-6-yl) acetamide) (Fig. 2) , a compound condensation product of Glycine and Adenine, which was considered as the most potential lead compound for further modification. Subsequently, the core hopping method was employed to search the fragment database for replacing the adenine part. Finally, the new structure Neo was discovered that has more strong affinity than ZINC01602230. The flowchart to show the process of finding the desired inhibitor is given in Fig. 2 , from which we can see that after the ZINC01602230 was screened out from the Zinc drug-like database, the core hopping method was used to optimize the core1 to core2 by means of searching the ZINC fragment database. The binding affinity between the NAG and the receptor was used as the filtering set. As a result, the compound with the top hits, Neo, was selected for further optimization. As shown in Fig. 1 , the rigid core2 fragment in Neo sticking out of the active pocket might not well adhere to the active pocket surface. To improve its binding affinity to the target protein, the bond C-N was cut off at the site shown in Fig. 2 . As a consequence of doing so, only the pyridine remained and the whole structural flexibility was enhanced so as to have the ability to stretch out to complement the surface of HA1 binding site. The new scaffold as a building block was further optimized through the second core hopping process by replacing the core3 with various fragments by searching the fragment database. Interestingly, the best substitute core4 also contains the same glycine as the terminal fragment on the other side of Neo or ZINC01602230. Subsequently, a series of compound candidates modified from the Neo structure were generated, and then the top ten compounds with the best binding affinity computed by Glide5 program [39, 42] were listed in Fig. 3 . As can be seen from Fig. 4 , the result obtained from the docking simulation has proved that the compound binding interactions with residues ARG227 and ASP92 were fully consistent with the previous report [57] . The structure of Neo6 complemented the shallow pocket of HA1 with the optimal conformation. The side chains of the key residues, such as Arg227, Pro143, Glu72 and Asp92 in protein, made a major contribution to the receptor-ligand binding affinity by forming H-bonds with the different heavy atoms (e.g. O, N) of the Neo6 (Fig. 4) . Besides the common H-bonds formed between the three residues (Arg227, Pro143, and Glu72) and the compound Neo6 as in [58] , the other two H-bonds were formed between the two nitrogen atoms of the new extensible fragment core4 and the oxygen atom of Asp92 residue. Consequently, compared with ZINC01602230, the binding affinity of Neo6 with the receptor was strengthened from 25.83 kcal/mol to 28.38 kcal/mol (Fig. 3 ). Furthermore, molecular dynamics simulations were performed for the inhibitor-complexed system HA1-Neo6 and the inhibitoruncomplexed system HA1, respectively. The root mean square deviation (RMSD) from initial conformation is a central criterion used to evaluate the difference of the protein system. The stability of a simulation system was evaluated based on its RMSD. The RMSD values for both Neo6-HA1 (green curve) and HA1 (red curve) versus the simulation time were illustrated in Fig. 5A , in which the RMSD for Neo6-HA1 system is a little smaller than that of HA1 system, indicating that the flexibility of HA1 was decreased after the Neo6 binding to HA1. In order to investigate the motions about the important residues interacted with the inhibitor in the binding site defined as loops (Loop1-Loop4) in Fig. 1 , the root mean square fluctuations (RMSF) for all the sidechain atoms of protein were calculated, as shown in Fig. 5B . The curves of RMSF associated with Loop1, Loop2, Loop3, and Loop4 are colored orange, light blue, dark blue, and maroon, respectively. It can be clearly seen from Fig. 5 that the fluctuating magnitudes of the four loops in HA1 are much larger than those in Neo6-HA1, clearly indicating that the receptor HA1 is more stable after binding with the ligand Neo6. The RMSD for all backbone atoms of the Neo6-HA1 system (green) and the HA1 system (red). (B) The RMSF for side-chain atoms of the Neo6-HA1 system (green) and the HA1 system (red). The curves associated with Loop1, Loop2, Loop3, and Loop4 are colored orange, light blue, dark blue, and maroon, respectively. doi:10.1371/journal.pone.0028111.g005 Accordingly, among the series of Neo compound candidates, Neo6 is anticipated to be a promising drug candidate for further experimental investigation to develop new and effective drug against influenza viruses.

Tamiflu-Resistant but HA-Mediated Cell-to-Cell Transmission through Apical Membranes of Cell-Associated Influenza Viruses

The infection of viruses to a neighboring cell is considered to be beneficial in terms of evasion from host anti-virus defense systems. There are two pathways for viral infection to “right next door”: one is the virus transmission through cell-cell fusion by forming syncytium without production of progeny virions, and the other is mediated by virions without virus diffusion, generally designated cell-to-cell transmission. Influenza viruses are believed to be transmitted as cell-free virus from infected cells to uninfected cells. Here, we demonstrated that influenza virus can utilize cell-to-cell transmission pathway through apical membranes, by handover of virions on the surface of an infected cell to adjacent host cells. Live cell imaging techniques showed that a recombinant influenza virus, in which the neuraminidase gene was replaced with the green fluorescence protein gene, spreads from an infected cell to adjacent cells forming infected cell clusters. This type of virus spreading requires HA activation by protease treatment. The cell-to-cell transmission was also blocked by amantadine, which inhibits the acidification of endosomes required for uncoating of influenza virus particles in endosomes, indicating that functional hemagglutinin and endosome acidification by M2 ion channel were essential for the cell-to-cell influenza virus transmission. Furthermore, in the cell-to-cell transmission of influenza virus, progeny virions could remain associated with the surface of infected cell even after budding, for the progeny virions to be passed on to adjacent uninfected cells. The evidence that cell-to-cell transmission occurs in influenza virus lead to the caution that local infection proceeds even when treated with neuraminidase inhibitors.

It is generally accepted that viruses, released as cell-free virions from an infected cell, transmit to distant cells and tissues. This spreading pathway contributes to wide-ranged diffusion of cell-free viruses. However, in this spreading pathway, viruses are exposed to host anti-virus defense systems. In contrast, direct infection to a neighboring cell is considered to be beneficial for the virus in terms of evasion from the host anti-virus defense. There are two typical manners in infection to ''right next door'': one is the virus transmission through cell-cell fusion by forming syncytium without production of progeny virions, and the other is mediated by virions without virus diffusion, generally designated cell-to-cell transmission [1, 2] . The cell-cell fusion infection pathway is characteristic for a variety of virus such as paramyxoviruses, herpesviruses, some retroviruses, and so on. For example in the case of measles virus belonging to Paramyxoviridae, infection is initiated by the interaction of the viral hemagglutinin glycoprotein with host cell surface receptors. The virus penetrates into the cell through membrane fusion mediated by the interaction of the fusion glycoprotein. In later stages of infection, newly synthesized glycoproteins accumulate at the cell membrane resulting in fusion of the infected cell with neighboring cells by producing syncytia. Thus, viruses can spread from cell to cell without producing cell-free virus particles. The examples of the cell-to-cell transmission are diverse, and these mechanisms are dependent on pairs of viruses and host cells. Vaccinia virus particles bound on the filopodium of an infected cell are repelled toward neighboring uninfected cells by the formation of filopodia using actin filament [3] . The filopodia direct viruses to uninfected cells. Immunotropic viruses including retroviruses utilize an immunological synapse, designed as virological synapses for the cell-to-cell transmission [4] [5] [6] [7] . Claudin-1 and occludin, components of tight junction, are involved in hepatitis C virus (HCV) entry through the cell-to-cell transmission [8, 9] . The cell-to-cell transmission through tight junction is also observed in other viruses which infect epithelial layers [10, 11] . These retroviruses and HCV remain on the surface of an infected cell even after budding. The uninfected cells adjacent to these infected cells can accept or take over viruses from the infected cell. Thus, the cell-to-cell transmission can be categorized into two manners based on the state of infecting viruses, either cell-free or cell-associated virions. Influenza virus, belonging to the family of Orthomyxoviridae, is one of the most serious zoonotic pathogens and causes seasonal epidemics or periodic pandemics among human beings around the world. The viral envelope consists of a lipid bilayer derived from cells that anchors three of viral transmembrane proteins, hemagglutinin (HA), neuraminidase (NA), and matrix protein 2 (M2). Influenza virus infection is initiated by the attachment of HA on virus particles to cell surface receptors containing sialic acids [12] . It has been known that the specific interaction between HA and sialic acid species is one of the determinants of the host range of influenza viruses [13] . Beside its role in the viral attachment, HA is also involved in intracellular fusion between viral envelope and host cell endosome membrane in the endocytotic pathway, by which the virus content is released inside the host cell [14] . The functional maturation of HA is mediated by the cleavage of HA into two disulfide-linked glycopolypeptides, HA1 and HA2 [15] , accomplished by trypsin or trypsin-like proteases derived from host cells [16] [17] [18] [19] . The membrane fusion is induced by a conformational change in the mature HA, which is triggered at low pH in the endosome, allowing viral ribonucleoprotein complexes to release into the cytoplasm [20, 21] . Thus, HA plays a critical role in initiation and progression of influenza virus infection. Influenza virus NA possesses the enzymatic activity that cleaves a-ketosidic linkages between terminal sialic acids and adjacent sugar residues of cellular glycoconjugates [22] . The sialidase activity of NA removes terminal sialic acid residues from HA and NA proteins as well as host cell surface glycoproteins. Since the terminal sialic acid of sialyloligosaccharides is critical for HA binding, the receptordestroying activity of NA serves to counter the receptor-binding activity of HA. It is quite likely that this activity contributes to prevention of successive superinfection of an infected cell [23] . In the absence of the functional sialidase activity, progeny virions aggregate on the cell surface due to the HA receptor-binding activity and can not be released [24, 25] . Thus, NA cleaves sialic acids from the cell surface and facilitates virus release from infected cells. However, it is not clear whether every progeny virion is released as cell-free virion to infect the uninfected cells after diffusion into the extracellular environment. Influenza viruses are generally transmitted as cell-free viruses from infected to uninfected cell but they may also infect through the cell-to-cell transmission, in particular during local lesion formation. Here, we examined whether influenza virus transmits from an infected cell to adjacent uninfected cells without virus release. Live cell imaging techniques showed that a recombinant influenza virus, in which the NA gene was replaced with the green fluorescence protein gene, spreads from an infected cell to adjacent cells forming infected cell clusters. Furthermore, progeny virions remain associated on the surface of infected cell even after budding, and then progeny virions could be passed to adjacent uninfected cells. To examine the transmission pathway of influenza virus, we performed immunofluorescence analyses by using anti-nucleoprotein (NP) polyclonal antibody. Influenza virus can form an infection center even in the presence of oseltamivir, a potent NA inhibitor (commercially known as Tamiflu) [26] [27] [28] . Oseltamivir at the concentration of 50 mg/ml completely prevented the release of progeny influenza viruses ( Figure 1A ). Noted that a large number of single fluorescent foci caused by initial infection markedly expanded and formed cell clusters consisting of 5-10 infected cells in an MDCK cell monolayer ( Figures 1B and S1 ), suggesting influenza virus can spread to some extent in the presence of oseltamivir. To verify that NA is not involved in this spreading, we generated an NA-deficient influenza virus by a reverse genetics method as described previously [29, 30] . The NAdeficient influenza virus contains a mutated NA segment, in which the NA coding region including a sialidase catalytic domain was replaced with the enhanced green fluorescent protein (EGFP) gene [29] . By this replacement, the NA activity is eliminated from the recombinant influenza virus, and EGFP can be utilized as a marker for viral infections. Immunofluorescence analyses demonstrated that the NA-deficient influenza virus also forms infected cell , culture supernatant was collected, and then its virus titer was determined by plaque assays. Each result was represented by a value relative to that in the absence of the drug. Error bars indicate standard deviation (s.d.) from 3 independent experiments. (B) Confluent MDCK cells were infected by wild-type influenza virus A/WSN/33 or NAdeficient influenza virus at MOI of 0.0001 in the presence or absence of 50 mg/ml oseltamivir phosphate. NA-deficient influenza virus was generated by reverse genetics as previously described [29] . After incubation at 37uC for 36 hours, immunofluorescence analyses were performed using anti-nucleoprotein (NP) polyclonal antibody and antirabbit IgG antibody conjugated to Alexa Fluor 568 (Invitrogen). Scale bar, 100 mm. doi:10.1371/journal.pone.0028178.g001 clusters similarly to those formed by wild-type influenza virus in the presence of oseltamivir ( Figure 1B) . The fluorescence pattern of NP overlapped with the localization of GFP derived from the EGFP gene of the NA-deficient influenza virus ( Figure S2 ). Thus, NA-deficient influenza virus can be used to investigate the NAindependent infection pathway of influenza virus. Next, we performed live cell imaging analyses to directly observe the infection time course of the NA-deficient influenza virus. The GFP fluorescence derived from the NA-deficient influenza virus first appeared in a single cell on an MDCK cell monolayer at 24 hours post infection. The virus started to spread from an infected cell to adjacent cells in 5-6 hours after the first appearance of a GFP-positive cell ( Figure 2 and Video S1). The spreading rate was clearly faster than the rate of cell divisions. The mean doubling time of uninfected MDCK cells was 20-24 hours under the condition employed here, and it is expected that the proliferation speed would be much slowly because infected MDCK cells were maintained in the serum-free medium and formed cell monolayer at the high cell density. These suggest that NA-deficient influenza viruses may infect adjacent cells through the cell-to-cell transmission mechanism without apparent production of cell-free virions. The cell-to-cell virus transmission pathway could be interpreted as one of viral evolving strategies to avoid neutralizing antibody responses [2, 31, 32] . Therefore, we examined the effect of neutralizing antibody on NA-deficient influenza virus. A polyclonal antibody with the neutralizing activity against influenza virus particles inhibited infection of cell-free viruses to less than 50% at the concentration of 0.03%, although the cell cluster formation was observed at the concentration less than 0.01%. On the other hand, the NA-independent transmission of the NA-deficient influenza virus was blocked only when neutralizing antibody was present at the concentration of 0.3% ( Figure 3 ). These results indicated that the NA-independent transmission of influenza viruses is less sensitive to the neutralizing antibody. Next, to investigate the mechanism of NA-independent transmission of influenza virus, we examined whether HA is involved in this transmission. In the absence of the NA activity, virus spreading from an infected cell to adjacent cells was dramatically suppressed by omission of trypsin, essential for maturation of HA, from the experimental condition ( Figure 4A ). The GFP fluorescence derived from NA-deficient influenza virus appeared in a single cell at 24 hours post infection. However, this virus did not spread, but rather disappeared during subsequent 24 hours (Video S2). These observations indicate that the NAindependent cell-to-cell transmission of influenza virus is dependent on HA maturation mediated by trypsin, as is the case for the general cell-free transmission of this virus. To clarify whether virus particles or viral RNP complexes are transmitted to adjacent cells, we examined the effect of amantadine on the cell-to-cell transmission of influenza virus. Amantadine inhibits the early step of uncoating of influenza virus RNP from virion in endosomes [33, 34] . For this study, other influenza virus strain, influenza virus A/Udorn/72, was used instead of influenza virus A/WSN/33 because influenza virus A/ WSN/33 is highly resistant to amantadine [35] . We confirmed that influenza virus A/Udorn/72 is sensitive to oseltamivir ( Figure S3 ) and could also spread via cell-to-cell transmission independent of the NA activity as did for influenza virus A/WSN/33 ( Figures 1B and 4B ). In the case of a single administration of amantadine, fluorescent foci derived from infected cells scattered, and the number of single foci was greatly decreased compared with that in the absence of the drugs. In contrast, a single administration of oseltamivir, fluorescent foci formed some clusters and expanded in a time-dependent manner ( Figure 4B ). This dissimilarity of inhibitory manner was caused by the difference of the sites of action between amantadine and oseltamivir. Amantadine inhibits the replication of influenza A virus by preventing the translocation of vRNP complexes from endosomes to the cytoplasm, whereas oseltamivir has no effects on viral replication itself but inhibits the release of cell-free virions from infected host cells. We investigated the inhibitory effect of amantadine on the cell-to-cell transmission of influenza viruses. The formation of infected cell clusters was observed with co-administration of amantadine and oseltamivir, as well as with a single administration of oseltamivir ( Figure 4B ). However, the quantitative analysis revealed that the size of infected cell clusters with the coadministration were decreased as compared to that with oseltamivir alone ( Figure 4C ). These observations indicated that the NA activity-independent cell-to-cell transmission of influenza virus was susceptible to the inhibitory effect of amantadine, suggesting that the cell-to-cell transmission undergoes through endocytosis but vRNP complex itself is not incorporated in the infected cells by adjacent cells. The virus transmission undergoes from infected to uninfected cells through either basolateral [36] [37] [38] or apical [39] [40] [41] [42] sides. In the case of influenza virus, cell-free progeny virions are released only from the apical surface of polarized epithelial cells [43] . This releasing polarity is achieved by directed transport of viral membrane proteins to the apical plasma membrane [44] . Indeed, that HA and NA glycoproteins are associated with lipid rafts, and the raft association has been implicated in apical transport [45, 46] . To determine whether or not the cell-to-cell transmission of the NA-deficient influenza virus occurs on the apical surface, we performed transwell assays in the presence of the neutralizing antibody to influenza A viruses. The neutralizing antibody was added to infected MDCK cell monolayer from apical or Immunofluorescence analyses were performed with cells infected with wild-type influenza virus at 18 hpi using anti-NP antibody and anti-rabbit IgG antibody conjugated to Alexa Fluor 488 (Invitrogen). GFP fluorescence derived from the recombinant virus was observed at 36 hpi. Scale bar, 100 mm. (C) The level of viral spreading was indicated in the graph by measuring NP and GFP derived from wild-type and NA-deficient virus, respectively. Five different microscope fields were taken randomly, and then the intensity of green color was analyzed with ImageJ NIH image processing software. Each result was represented by a value relative to that in the absence of neutralizing antibodies. Error bars indicate s.d. from 3 independent experiments. doi:10.1371/journal.pone.0028178.g003 basolateral side, and the inhibitory effect on the spread of GFP fluorescence derived from the recombinant virus was examined. Addition of high concentrations of the neutralizing antibody from the apical side blocked the cell-to-cell transmission of the NAdeficient influenza virus, whereas the addition from the basolateral side had no effect ( Figure 5 ). These observations indicated that the polarity in the influenza virus budding in the cell-to-cell transmission pathway is apical. Previous report showed that influenza viruses were refractory to superinfection with a second cell-free virus [23] . In the case of the cell-to-cell transmission of influenza virus in the presence of oseltamivir, it is possible that a progeny virion is temporarily bridged by HA between an infected cell and adjacent uninfected cells, since viruses can not be released from infected cell surface due to the inhibition of the NA activity by oseltamivir. The cellassociated progeny virion may have an opportunity to re-infect the previously infected cell, compared to a cell-free progeny virion in the general spreading. Thus, we examined whether influenza viruses can infect the cell which had already been infected, using ts53 mutant and wild-type influenza virus A/WSN/33. ts53 virus has a substitution mutation from U to C at the nucleotide position of 701 in the PA gene. This substitution introduces an amino acid change from wild-type Leu 226 to Pro 226 and gives a defect in the viral genome replication process [47, 48] . At first, cells were infected with ts53 virus at moi of 10, and after incubation for 0, 2, 4, 6, and 8 hours, cells were superinfected with wild-type virus at moi of 10. The amount of segment 3 viral RNA (vRNA) encoding PA was determined quantitatively by RT-PCR. Then, using a mutated primer for PCR, we could introduce a Stu I site only in the PCR products derived from the wild-type sequence ( Figure 6A ). Thus, DNA fragments amplified from the wild-type and ts53 could be distinguished by Stu I digestion. The digested DNA fragments containing 220 and 199 base pairs derived from ts53 and wildtype, respectively, were separated through PAGE. After 6 hours or later post infection, re-infection with the second challenging virus hardly occurs in the absence of oseltamivir. However, in the presence of oseltamivir, appearance of wild-type fragment suggests that the re-infection had occurred ( Figure 6B ). The result indicates that progeny virus particles remain on the surface of infected cell even after budding, and can infect the cell previously infected, as well as uninfected cells adjacent to the infected cell, when oseltamivir is present. With the except for the virus which spreads through the cell-cell fusion transmission, virus infection is initiated by the binding of cell-free virions to their host cells. Recently, the virus transmission mechanism from an infected cell to adjacent cells without virus diffusion into the extracellular environment is highlighted from the aspect of its significance in virus spreading in the presence of antibodies [1, 2] . This antibody-insensitive pathway is often called cell-to-cell transmission [2] . The cell-to-cell transmission may be categorized into two pathways, i.e., transmission of cell-free virions to adjacent uninfected cells, and transmission of progeny virions associated on the surface of an infected cell even after budding through narrow synaptic space between an infected cell and adjacent uninfected cells. As an example of the former mechanism, cell-free vaccinia virus particles associated with the filopodium of an infected cell are repelled toward neighboring uninfected cells by inducing the formation of actin filament [3] . Several cases have been reported for the latter mechanism: Immunotropic viruses including retroviruses utilize the immunological synapses [4] [5] [6] [7] . Immune cells are not constitutively polarized, but contain the machinery that directs their secretory apparatus towards a cell that is involved in an immunological synapse. This machinery can be subverted by retroviruses containing human immunodeficiency virus (HIV). An HIV-infected cell can polarize viral budding towards a target cell expressing receptor through a structure called a virological synapse. Virions bud from an infected cell into a synaptic cleft, from which they fuse with the target-cell plasma membrane [49] [50] [51] [52] . The progeny virions of HCV are trapped between infected and uninfected cell membranes at the tight junction. Using Claudin-1 known as a component of the tight junction and one of the entry factors of HCV [8] , virions fuse with and penetrate uninfected target cells [31] . Therefore, HCV may acquire the ability to spread within polarized liver epithelium. Thus, the cell-to-cell transmission certainly plays significant roles for the dissemination of several enveloped viruses. However, the cell-to-cell transmission of influenza virus has not been discussed well. Here, we have shown that influenza virus spreads by forming infected cell clusters even in the presence of an NA inhibitor. Live cell imaging clearly showed that influenza virus lacking the NA activity spreads from an infected cell to adjacent cells through the cell-to-cell transmission mechanism (Figure 2 ). This was also the case for wild-type influenza virus during early phases of infection ( Figure 4B ). In the cell-to-cell transmission of influenza virus, progeny virions could remain associated with the surface of infected cell even after budding, and then these progeny virions can be passed on to adjacent uninfected cells. We showed that the cell-to-cell transmission of the NA-deficient influenza virus depends on functional HA. The viral spreading was dramatically suppressed without HA activation by trypsin treatment ( Figure 4A) . Moreover, the cell-to-cell transmission was also blocked by amantadine, which inhibits the acidification of endosomes required for uncoating of influenza virus particles in endosomes [33, 34] . These findings indicate that functional HA and endosome acidification by M2 ion channel are required for the cell-to-cell influenza virus transmission, thereby allowing viruses to enter the adjacent cells through the endocytotic pathway ( Figure 4) . Our findings showed that the NA-deficient influenza virus is not diffused into the extracellular environment. The viral spreading in the absence of oseltamivir appears to be much faster compared to the viral spreading in the presence of the drug, suggesting that NA could be involved in determination of spreading speed ( Figure 4B ). The NA activity prevented progeny virions from entering cells which virus came from (Figure 6 ), implying that progeny virus particles should be transmitted to adjacent uninfected cells. The cell-to-cell transmission started in early phase of infection, and the virus spread through diffusion of cell-free viruses ( Figure 4B ). Indeed, it was reported that the cell-to-cell transmission is a rapid spreading pathway in the case of vaccinia virus [3] . Vaccinia virus induces a blocking mechanism of superinfection and thereby infects to adjacent uninfected cells efficiently. In early phases of vaccinia virus infection, viral proteins A33 and A36 are expressed at the infected cell surface. Once cell-free virus particles contact the filopodium, the A33/A36 complex induces the formation of actin filament, which causes this superinfected virion to be repelled toward uninfected cells [3] . Influenza viruses can re-infect the cells previously infected in the presence of oseltamivir (Figure 6 ), suggesting that a progeny virion may be bridged by HA between infected and adjacent uninfected cells temporarily. Thus, in the case of the cell-to-cell transmission of influenza virus, we propose that progeny virions associated with the surface of infected cells even after budding are directed to adjacent uninfected cells. The cell-to-cell transmission mechanism of influenza virus is distinctly different from that of vaccinia virus in the infecting virus status: Infected cell-associated virions and cell-free virions are involved in the cell-to-cell transmission of influenza virus and vaccinia virus, respectively. The strategy for influenza virus appears to be similar to that for HCV. HCV progeny virions budded from an infected cell are trapped between infected and uninfected adjacent cell membranes at the tight junction. HCV virions then, enter into adjacent cells through endocytosis and low pH-dependent membrane fusion using Claudin-1 [8] . The cell-to-cell transmission of influenza virus also required functional HA and endosome acidification by M2 ion channel. However, it has not been reported that HCV has a gene encoding a receptor destroying enzyme similar to NA of influenza virus. We speculated that HCV progeny particles are bridged between infected and adjacent uninfected cells temporarily like influenza virus in the presence of oseltamivir. Progeny influenza virus particles could be transmitted to adjacent uninfected cells efficiently in the presence of the NA activity, suggesting that the cell-to-cell transmission of influenza virus is more strategic than that of HCV. Our findings raise an interesting question as to what is the biological significance of cell-to-cell transmission for influenza virus infection in vivo. Until now, it had been believed that influenza virus was released from infected cells as cell-free virions and then spread from cell to cell as well as from organism to organism. The transmission mode by cell-free virions undergoes the extremely highspeed of its diffusion and causes epidemic or pandemic infection. The tropism in an infected animal body is generally restricted to respiratory tract or lung and its periphery, and the requirement of a trypsin-like protease has been generally described for the reason of the restriction. It is possible that the cell-to-cell transmission mode may play a significant role for the virus spreading inside of organism, although cell-free influenza virions are causative of highspeed spreading. At the least, the limited but distinct level of infection followed by replication could provide some opportunity to generate influenza virus variants. It is an open question whether the cell-to-cell transmission mode is involved in the pathogenesis caused by influenza virus infection in vivo. The existence of cell-to-cell transmission pathway gives a caution when NA inhibitors are used, because NA inhibitors may not be sufficient to completely block the spread of influenza Figure 6 . Influenza viruses can not re-infect previously infected cells. (A) A method for determination of the amount of segment 3 genome derived from ts53 and wild-type. Total RNA was reverse-transcribed with the primer PA-895-rev, which is complementary to the segment 3 positivesense RNA. The cDNA was amplified by PCR using primers, PA-895-rev and PA-695-cut partially corresponding to segment 3 positive sense RNA between the nucleotide sequence positions 678 to 700 except for 696 and 697, which are shown in red letters. Since segment 3 of ts53 has a substitution mutation from U to C at the nucleotide position of 701, the PCR product derived from wild-type could be digested by Stu I but not that from ts53. Then, PCR products were digested with Stu I and separated through 8% PAGE. (B) Detection of the genome of the segment 3 derived from ts53 or wild-type. At 3 hours post superinfection of wild-type virus, total RNA was extracted, and semi-quantitative RT-PCR was performed. Subsequently, the amplified DNA products were digested with Stu I and separated through 8% PAGE. Large and small fragments derived from ts53 and wild-type viruses were 220 and 199 base pairs, respectively. The relative amount of wild-type segment 3 to that at 0 hour in the absence of oseltamivir phosphate was shown in the graph. Error bars indicate S.D. from 3 independent experiments. White bar, in the absence of oseltamivir phosphate; black bar, in the presence of oseltamivir phosphate. doi:10.1371/journal.pone.0028178.g006 virus in local microenvironments. Since this cell-to-cell transmission pathway exists, development of antiviral therapeutic strategies in addition to NA inhibitors is highly recommended. Madin-Darby canine kidney (MDCK) cells were kindly gifted by A. Ishihama (Hosei University), and maintained in minimal essential medium (MEM) (Nissui) containing 10% fetal bovine serum. Human embryonic kidney 293T cells were kindly gifted by Y. Kawaoka (University of Tokyo), and maintained in Dulbecco modified Eagle medium (DMEM) (Nissui) supplemented with 10% fetal bovine serum. Influenza virus A/Udorn/72 was grown in allantoic sacs of 11 day-old embryonated eggs (MIYAKE HATCHERY). Wild-type influenza virus A/WSN/33 and ts53 mutant were used after single-plaque isolation. MDCK cells were infected with influenza virus A/WSN/33 or ts53 at a multiplicity of infection (MOI) of 0.1 PFU/cell, and incubated at 37uC and 34uC, respectively. After incubation for 24 h, the culture fluid was harvested and centrifuged at 1,7006 g for 10 min. The virus suspension was stored at 280uC until use. The production of rabbit polyclonal anti-NP antibody was described previously [53] , and this antibody was used as a primary antibody for indirect immunofluorescence assay. A goat antirabbit IgG antibody conjugated to Alexa Fluor 488 or Alexa Fluor 568 was purchased from Invitrogen and used as a secondary antibody for indirect immunofluorescence assay. A polyclonal antibody against influenza A virus was obtained from 2-month-old female rabbit immunized with 250 mg of purified virions of influenza virus strain A/Puerto Rico/8/34 [54] . The generation of antibodies was boosted three times and used as neutralizing antibodies to block the influenza virus infection. MDCK cells were infected with influenza virus A/WSN/33 at a multiplicity of infection (MOI) of 0.001 PFU per cell. After virus adsorption at 37uC for 1 hour, the cells were washed with serumfree MEM and incubated at 37uC with maintenance medium (MEM containing vitamins and 0.1% BSA) containing oseltamivir. At 48 hours post infection (hpi), culture supernatant was collected, and then its viral titer was determined by plaque assays. An NA-deficient influenza virus possessing the terminal sequences of NA segment but lacking the NA coding region, which was replaced with enhanced green fluorescent protein (EGFP) gene, was generated by reverse genetics as described previously [29, 30] . For reverse genetics, we used plasmids containing cDNAs of the influenza virus A/WSN/33 viral genome under the control of the human RNA polymerase I promoter (referred to as Pol I plasmids). Briefly, 293T cells were transfected with seven Pol I plasmids for production of all vRNA segments of influenza virus A/WSN/33 and one for the mutant NA vRNA segment containing EGFP ORF, together with protein expression vectors for PB2, PB1, PA, and NP controlled by the chicken b-actin promoter (pCAGGS). TransIT-293 (Mirus) was used for transfection. At 24 hours post transfection, recombinant viruses were harvested from the cell surface using bacterial NA derived from Clostridium perfringens (sigma). MDCK cells were infected with harvested recombinant viruses treated with N-tosyl-L-phenyl-alanine chloromethyl ketone (TPCK)-trypsin (1 mg/ml). After confirmation of GFP fluorescence derived from amplified recombinant virus genomes at 48 hours after infection, the recombinant viruses on the cell surface were collected using bacterial NA. The viral titer of recombinant viruses was determined by counting the number of infected foci using a fluorescence microscopy (Carl Zeiss). Cells on coverslips were fixed with 4% paraformaldehyde in phosphate-buffered saline (PBS) for 10 min and permeabilized with 0.2% NP-40 in PBS. The coverslips were soaked in 1% bovine serum albumin in PBS, and then incubated at room temperature for 1 hour with a primary antibody. After being washed twice with PBS, the coverslips were incubated at room temperature for 1 hour with a secondary antibody. The coverslips were then incubated at room temperature for 5 min with 3 mM 49,69-diamidino-2phenylindole (DAPI) and finally mounted on glass plates, and cells were observed under the fluorescence microscope. Living cells were analyzed using BioStation ID system (GE Healthcare). Confluent MDCK cells were infected with the NAdeficient influenza virus at the multiplicity of infection (MOI) of 0.0001 in the presence or absence of 1 mg/ml TPCK-trypsin. At 24 hours post infection, culture dishes containing infected cells were set into the chamber of BioStaion ID system, which was maintained at 37uC under 5% CO 2 and 95% humidity. Then, images were acquired during next 24 hours at interval with 1 hour. The excitation wavelength was controlled by a manual filter wheel equipped with filters suitable for enhanced green fluorescence protein (EGFP). Confluent MDCK cell monolayer was prepared on transwell inserts (BD Falcon, pore size 0.4 mm) and infected with the NAdeficient influenza virus at MOI of 0.0001. After virus adsorption at 37uC for 1 hour, the cell monolayer was washed with serum-free MEM, and maintenance medium was added into both sides within the transwells. The neutralizing antibody to influenza A virus was added into the inside or the outside of transwell inserts with the maintenance medium. Subsequently, cells were incubated at 37uC for 36 hours followed by analyses using the fluorescence microscopy. ts53 virus has a substitution mutation from U to C at the nucleotide position of 701 in the PA gene. This substitution introduces an amino acid change from wild-type Leu 226 to Pro 226 and gives a defect in the viral genome replication process [48] . However, under the permissive temperature, the level of viral genome replication is no difference between wild-type and ts53 [47] . To discriminate the genome of wild-type and that of ts53, total RNA was reverse-transcribed by reverse transcriptase (TOYOBO) with PA-895-rev (59-TTAATTTTAAGGCATC-CATCAGCAGG-39), which is complementary to the segment 3 positive sense RNA. The cDNA was amplified by PCR using primers, PA-895-rev and PA-695-cut (59-TCTCCCGCCA-AACTTCTCAGGCC-39) partially corresponding to segment 3 positive sense RNA between nucleotide sequence positions 678 to 700 except for nucleotide positions 696 and 697. Since segment 3 of ts53 has a substitution mutation from U to C at the nucleotide position of 701, the PCR product derived from wild-type was digested by Stu I but not that from ts53. After PCR reactions, PCR products were digested with Stu I and separated through PAGE. Large and small fragments derived from ts53 and wild-type viruses were 220 and 199 base pairs, respectively. DNA was stained with GelRed (BIOTIUM) and visualized by UV illumination. Figure S1 Formation of cell cluster caused by initial infection. MDCK cells were infected with influenza virus A/ WSN/33 at moi of 0.0003 in the presence or absence of 50 mg/ml oseltamivir phosphate. After incubation for 8 and 24 h, immunofluorescence analyses were performed using anti-NP antibody and anti-rabbit IgG antibody conjugated to Alexa Fluor 488 (Invitrogen). Nuclear DAPI and viral NP staining patterns are shown in blue and green, respectively. Enlarged views are shown in red borders. Scale bar, 100 mm. (TIF) Figure S2 The expression of GFP derived from NAdeficient influenza virus overlapped with the localization of NP. MDCK cells were infected with NA-deficient influenza viruses at MOI of 0.0001. After incubation at 37uC for 48 hours, immunofluorescence analyses were performed using anti-NP antibody. Scale bar, 100 mm. (TIF) Figure S3 Influenza virus A/Udorn/72 was sensitive to oseltamivir. MDCK cells were infected with influenza virus A/ Udorn/72 at a MOI of 0.001 PFU per cell. At 36 hpi, the culture supernatant was collected, and then its virus titer was determined by plaque assays. Each result was represented by a value relative to that in the absence of the drug. Error bars indicate s.d. from 3 independent experiments. (TIF) Video S1 NA-deficient influenza virus spreads through cell-to-cell transmission. Confluent MDCK cells were infected with NA-deficient influenza virus at MOI of 0.0001 in the presence of trypsin. After incubation at 37uC for 24 hours, a single GFP-positive cell and its vicinity were traced it during the period from 24 hpi to 48 hpi at interval of 1 hour. Live cell imaging data analyses was performed by Biostation ID (GE healthcare). Scale bar, 50 mm. Video S2 NA-deficient influenza virus does not spread in the absence of trypsin. Confluent MDCK cells were infected with the NA-deficient influenza virus at MOI of 0.0001 in the absence of trypsin. After incubation at 37uC for 24 hours, a single GFP-positive cell was detected, and then this cell and neighborhood cells was traced during the period from 24 hpi to 48 hpi at interval of 1 hour. Live cell imaging data analyses were performed by Biostation ID (GE healthcare). Scale bar, 50 mm. (MOV)

Persistent Expression of Hepatitis C Virus Non-Structural Proteins Leads to Increased Autophagy and Mitochondrial Injury in Human Hepatoma Cells

HCV infection is a major cause of chronic liver disease and liver cancer in the United States. To address the pathogenesis caused by HCV infection, recent studies have focused on the direct cytopathic effects of individual HCV proteins, with the objective of identifying their specific roles in the overall pathogenesis. However, this approach precludes examination of the possible interactions between different HCV proteins and organelles. To obtain a better understanding of the various cytopathic effects of and cellular responses to HCV proteins, we used human hepatoma cells constitutively replicating HCV RNA encoding either the full-length polyprotein or the non-structural proteins, or cells constitutively expressing the structural protein core, to model the state of persistent HCV infection and examined the combination of various HCV proteins in cellular pathogenesis. Increased reactive oxygen species (ROS) generation in the mitochondria, mitochondrial injury and degeneration, and increased lipid accumulation were common among all HCV protein-expressing cells regardless of whether they expressed the structural or non-structural proteins. Expression of the non-structural proteins also led to increased oxidative stress in the cytosol, membrane blebbing in the endoplasmic reticulum, and accumulation of autophagocytic vacuoles. Alterations of cellular redox state, on the other hand, significantly changed the level of autophagy, suggesting a direct link between oxidative stress and HCV-mediated activation of autophagy. With the wide-spread cytopathic effects, cells with the full-length HCV polyprotein showed a modest antioxidant response and exhibited a significant increase in population doubling time and a concomitant decrease in cyclin D1. In contrast, cells expressing the non-structural proteins were able to launch a vigorous antioxidant response with up-regulation of antioxidant enzymes. The population doubling time and cyclin D1 level were also comparable to that of control cells. Finally, the cytopathic effects of core protein appeared to focus on the mitochondria without remarkable disturbances in the cytosol.

Hepatitis C virus (HCV) is an enveloped, positive, singlestranded RNA virus in the family of Flaviviridae [1] . The linear, non-segmented HCV genome of 9.6 kb encodes a polyprotein that undergoes post-translational cleavage by cellular and viral proteases to yield at least 10 mature proteins [2] [3] [4] . HCV infection is a major cause of chronic liver disease and is the major cause of liver cancer in the United States. HCV produces a chronic infection in 50-80% of infected patients; among them, roughly 20% will eventually develop liver cirrhosis. It is widely accepted that insufficient host immune response in eliminating HCV leads to persistent infection and the eventual development of liver diseases [4] [5] [6] . Interferon-a and ribavirin treatments have been prescribed either to stimulate immune response for clearance of viruses or to disrupt viral replication. However, high toxicity and low efficacy toward the two most prevalent HCV subtypes, 1a and 1b, in the US has been a barrier to effective eradication of persistent HCV infections [7] . To address the pathogenesis caused by HCV infection, recent studies have begun to focus on direct cytopathic effects. HCV proteins associate with different subcellular structures, including mitochondria, endoplasmic reticulum (ER), and lipid droplets, to facilitate replication and assembly of viral particles [2] . These associations lead to alterations of the integrity and functions of organelles. HCV-mediated oxidative stress is commonly observed and is achieved by increasing reactive oxygen and nitrogen species (ROS and RNS) or by altering cellular antioxidant capacities [8] [9] [10] [11] . In particular, HCV core proteins are shown to be closely associated with the mitochondria and cause increases in ROS and RNS production and lipid peroxidation [11] [12] [13] [14] , reduction in GSH and NADPH concentrations, reduction in mitochondrial complex I activities, and increase in mitochondrial Ca +2 uptake, which ultimately disrupts mitochondrial membrane permeability and leads to mitochondrial dysfunction [14, 15] . HCV nonstructural proteins have also been implicated in disturbing the redox balance and altering antioxidant enzyme levels [16, 17] . Specifically, NS5A is shown to up-regulate Mn superoxide dismutase (MnSOD) through AP1 transcription factor in the p38 MAPK and JNK signaling pathways [18, 19] . Additional studies showed the involvement of NS5A in ER stress and disturbance of intracellular Ca +2 homeostasis, which leads to increased mitochondrial ROS production and altered mitochondrial function [18, 20] . Because of the relationship between chronic HCV infection and the development of hepatocellular carcinoma, studies have also been carried out to identify HCV proteins that may be responsible for the hepatocarcinogenesis. For example, the HCV core protein has been shown to promote immortalization of primary human hepatocytes [21] , whereas the non-structural proteins NS3 and NS4B have been shown to transform NIH 3T3 cells either individually or in combination with Ha-ras [22, 23] . Most studies have focused on the direct cytopathic effects of individual HCV proteins, with the objective of identifying their specific roles in the overall pathogenesis. However, this approach precludes examination of the possible interactions between different HCV proteins and organelles. We hypothesize that different components of HCV polyprotein, depending on their subcellular distributions, cause different types of cytopathic effects and elicit different types of responses in different subcellular compartments. To obtain a better understanding of the various cytopathic effects of and cellular responses to HCV proteins, we used hepatoma cells constitutively expressing the HCV genomelength replicon, the subgenomic replicon, or the core protein to model the state of persistent HCV infection and took a comparative approach to dissecting the role of various HCV proteins in cellular pathogenesis in this study. Based on the SEAP activities measured from multiple sampling, the protein expression levels of HCV genome-length and subgenomic replicons were comparable ( Figure S1A ). Core protein expression levels, on the other hand, suggested that Core-on cells produced higher levels of core protein than the genome-length replicon cells ( Figure S1B ). To identify the subcellular location of HCV proteins in genome-length replicon, subgenomic replicon, and Core-on cells, immunogold EM was carried out with antibodies against core, NS5A, and NS5B proteins. The majority of core, NS5A, and NS5B proteins were located in the mitochondria and the ER with minor differences in their distribution within these organelles ( Figure 1A and Table S1 ). Positive signals were also observed in the nucleus, lipid droplets, and the Golgi. NS5A signals were observed in autophagocytic vacuoles in 25% of the electron micrographs examined, with an average signal intensity of 3-6 gold particles per autophagocytic vacuole (Table S1 ). To identify ultrastructural changes caused by the presence of HCV proteins, electron microscopic analyses were carried out. The study results revealed mitochondrial injury as a common defect among genome-length replicon, subgenomic replicon, and Core-on cells with enlarged mitochondria and focal loss of cristae ( Figure 1B ). Consistent with this observation, measurement of the cross sectional area of mitochondria directly from electron micrographs showed a significant increase in mitochondrial sizes ( Figure 2A ). The data indicates mitochondrial degeneration and the possible loss of mitochondria. Consequently, the number of mitochondria per cell was significantly reduced ( Figure 2B ). In addition to mitochondrial defects, accumulation of lipid droplets was common among all three HCV protein-expressing cell lines ( Figure 1B and Figure S2 ). Genome-length and subgenomic replicon cells also showed focally dilated rough endoplasmic reticulum (RER) and prominent presence of autophagocytic vacuoles ( Figures 1B and 3A ). Increased oxidative damage is the likely culprit for the observed mitochondrial degeneration and reduced mitochondrial number in HCV protein-expressing cells. To determine the extent of ROS generation in the mitochondria, MitoSOX was used for live cell staining. The results showed a significant increase in ROS production in the mitochondria across all three cell lines ( Figures 2C and 2D ). After adjusting to the average number of mitochondria per cell, the level of MitoSOX intensity was similar among genome-length replicon, subgenomic replicon, and Coreon cells. Cytosolic and mitochondrial aconitases (ACO1 and ACO2) are very sensitive to inactivation by reactive oxygen and nitrogen species, and reduction in aconitase protein levels and enzyme activities have been used as indicators for increased oxidative stress in their respective subcellular compartments [24, 25] . Our studies showed that ACO1 existed at low levels, whereas ACO2 was barely detectable by western blot analysis, in Huh7 and HCV protein-expressing cells. ACO1 protein level was reduced by 18% in cells with the genome-length replicon ( Figure S3A ), and total aconitase activity was reduced by 5-23% ( Figure S3B ) in genome-length and subgenomic replicon cells. The reduction in total aconitase activity most likely reflected increased oxidative stress in the cytosol because the majority of aconitase in Huh7 cells was in cytosolic form. From ultrastructure analyses, we consistently observed the presence of autophagocytic vacuoles and primary autophagocytic vesicles ( Figures 1B, 3A , and 3B) in genome-length replicon and subgenomic replicon cells and, to a lesser extent, in Core-on cells. To confirm the initial observation, immunocytochemical and western blot analyses were carried out to determine the status of LC3, which is an integral part of the autophagosome membrane. LC3-positive punctuate structures in the cytoplasm were prominent in genome-length replicon and subgenomic replicon cells ( Figure 3C ), but they were only present at a very low level in Coreon cells. Consistent with this result, western blot analyses showed a marked increase of LC3-I and LC3-II in genome-length replicon and subgenomic replicon cells ( Figure 3D ). However, the ratios between LC3-I and II were not changed. In contrast to the prominent accumulation of autophagocytic vacuoles in genomelength and subgenomic replicon cells, autophagosomes were not detected in HCV transgenic mice expressing the full-length HCV polyprotein (data not shown). To determine if oxidative stress played a role in HCV-mediated activation of autophagy, cellular redox state was altered by either enhancing the cellular antioxidant capacity through dual overex-pression of superoxide dismutase (SOD) and catalase (CAT), or increasing cellular oxidative stress by xanthine/xanthine oxidase (X/XO) treatment. Changes in total LC3 and LC3-II levels were used as indicators of autophagy. Significant reduction of total LC3 and LC3-II levels was observed in subgenomic replicon, Core-on, and Core-off cells with overexpression of CuZnSOD/cytosolic catalase (SOD1/cCAT) or MnSOD/mitochondrial catalase (SOD2/mCAT) ( Figure 4A ). Although similar trends were observed in Huh7 cells, the extent of reduction did not reach a significant level. Despite comparable levels of CuZnSOD, MnSOD, and catalase expression ( Figure S4 ), LC3 levels were not altered in genome-length replicon cells. In contrast, X/XO treatment significantly increased total LC3 levels in genomic replicon, subgenomic replicon, and Core-off cells ( Figure 4B , left panel). However, LC3-II levels were only significantly increased in genomic replicon and Core-off cells ( Figure 4B , right panel). No significant change in LC3-II level was observed in Huh7, subgenomic replicon, and Core-on cells. Furthermore, addition of CAT to the X/XO treatment diminished the increase of total LC3 in genome-length and subgenomic cells ( Figure 4B ). The data suggest that the enhanced autophagy from X/XO treatment is mediated through H 2 O 2 in genome-length and subgenomic replicon cells. To find out whether increased mitochondrial stress and ER stress led to changes in antioxidant profiles, western blot analyses were carried out to determine the protein levels of CuZnSOD, MnSOD, peroxiredoxin 1 (PRDX1), peroxiredoxin 3 (PRDX3), thioredoxin 1 (TRX1), and thioredoxin 2 (TRX2). Among them, CuZnSOD, PRDX1, and TRX1 are cytosolic proteins, and MnSOD, PRDX3, and TRX2 are mitochondrial proteins. The sulfonylated peroxiredoxins (PRDX-SO3), which are the end products of irreversible oxidation of the active site cysteine in peroxiredoxins, were also monitored and served as indicators for the redox state in HCV protein-expressing cells ( Figure S5C ). Significant increases in the protein levels of CuZnSOD, MnSOD, PRDX1, PRDX3, TRX1, and TRX2 were consistently observed in the subgenomic replicon cells, and the extent of increase ranges from 1.5-to 4.9-fold ( Figure 5 ). The genome-length replicon cells showed a modest (1.8-fold) but significant increase in TRX2, and a marginal decrease in CuZnSOD (p = 0.0734) and PRDX3 (p = 0.0739) ( Figure 5 ). No remarkable changes were observed between Core-on and Core-off cells. The ratios of PRDX1-SO3/ PRDX1 and PRDX3-SO3/PRDX3 were also used to gauge the redox state in the cytosol and the mitochondria, respectively, and no significant changes were observed across all cell lines analyzed ( Figure S5A and S5B). To determine if ultrastructural changes and increased mitochondrial ROS production affect cell proliferation and survival, population doubling time was determined during the exponential phase of cell growth. Genome-length replicon and Core-on cells showed a 41% and 11% increase in population doubling time, respectively, without a significant increase in cell death during the 72-hr period when cell number increase was monitored ( Figure 6A ). On the other hand, subgenomic replicon cells had a comparable population doubling time to that of Huh7 and Coreoff cells. To determine if cell cycle regulation is affected in HCV protein-expressing cells, in-cell westerns were carried out to determine total cyclin D1 levels. Consistent with increased population doubling time in genome-length replicon cells, cyclin D1 level was decreased ( Figure 6B ). There was also a trend in decreased cyclin D1 levels in Core-on cells. In this study, we showed that the most dominant subcellular location for HCV core, NS5A, and NS5B proteins was in the mitochondria, followed by the ER and the Golgi. Consequently, expression of HCV core or non-structural proteins led to increased ROS generation in the mitochondria, which most likely contributed to mitochondrial injury and degeneration. Expression of non-structural proteins also led to membrane blebbing in the ER and accumulation of autophagocytic vacuoles. Despite these changes, only cells with the subgenomic replicon (i.e. cells expressing the non-structural proteins) showed consistent upregulation of mitochondrial and cytosolic antioxidant enzymes. Cells with the genome-length replicon, on the other hand, did not have a robust antioxidant response and showed prolonged population doubling time and a decrease in cyclin D1 expression. Apart from the shared mitochondrial phenotype, increased lipid accumulation was also common among all three HCV proteinexpressing cell lines. Subcellular locations of HCV proteins have provided important clues to the site of viral genome replication and assembly, as well as the cytopathic effects caused by HCV infection. The ultrastructure study with immunogold labeling showed a distinct localization of core, NS5A, and NS5B proteins in the mitochondria and the ER. Significant core, NS5A, and NS5B distributions were observed in other subcellular compartments, including the Golgi apparatus, the nucleus, and lipid droplets. The study results are in general agreement with previously published data [26] [27] [28] [29] [30] [31] . However, the inner membrane location of core protein was unexpected because previous studies using proteinase K digestion suggested association of core protein with the mitochondrial outer membrane and the mitochondrial-associated membrane compartment [15, 28] . The discrepancy could be due to the compromised mitochondrial membrane integrity in genome-length replicon and Core-on cells, or due to artifacts from sample preparations for EM analysis. The association of core protein with lipid droplets also appeared to be lower than previous study results, and the discrepancy could be due to differences in the antibodies used. A significant number of studies have focused on the role of HCV core protein in mitochondrial dysfunction [11, 14, 15, 20, 26, 32] . In this study, cells expressing only the nonstructural proteins were also shown to have suffered from mitochondrial damage, and the extent of damage, in terms of mitochondrial size and number, was comparable to that of cells expressing the full-length polyprotein or core protein (Figures 2A and 2B ). In addition, the level of ROS production in the mitochondria, as detected by MitoSOX, was also comparable among the three HCV protein-expressing cell lines after adjusting for total mitochondria (Figures 2C and 2D) . The data suggest that multiple HCV proteins are capable of entering the mitochondria and cause increased oxidative stress, although the underlying mechanism may not be the same for each protein. The core protein-expressing cells had the lowest number of mitochondria per cell ( Figure 2B ). The data implies that Core-on cells may suffer from the most severe mitochondrial loss among the three HCV protein-expressing cell lines. Whether this is due to the higher level of core protein expression in Core-on cells ( Figure S1A ) will need to be examined with other approaches. Besides increased ROS production in the mitochondria of all three cell lines analyzed, our data on aconitase protein levels and enzyme activities also suggested an increased ROS production in the cytosol in genome-length and subgenomic replicon cells. The reduction in ACO1 protein level in genome-length replicon cells is reminiscent to that observed in CuZnSOD deficient mice [25] , in which heightened state of oxidative stress in the cytosol may lead to increased degradation of irreversibly inactivated aconitase. Subgenomic replicon cells had reduced aconitase activity without a significant reduction in the protein level ( Figures S3A and S3B) . The data suggests that ACO1 may be reversibly inactivated due to increased oxidative stress in the cytosol. Taken together, our study results suggest that HCV non-structural proteins are capable of causing increased oxidative stress in both the mitochondrial and the cytosolic compartments, whereas the effects of core protein is mainly limited to the mitochondria. Cells respond to a variety of internal stress or cellular damages by forming autophagosomes to degrade damaged components and to recycle usable elements for future biosynthesis [33] . However, some RNA viruses, most notably, poliovirus, dengue virus, mouse hepatitis virus, and foot-and-mouth disease virus develop the abilities to overcome such cellular defense by hijacking the autophagocytic pathway to facilitate viral replication [34] [35] [36] [37] . Multiple studies in recent years have shown that HCV infection of the human hepatoma cell line Huh7 and immortalized human hepatocytes induce autophagocytic vacuoles [38, 39] and that autophagy machinery is required for the initiation of HCV replication [40] and the production of infectious particles [41] . HCV induces the accumulation of autophagosomes via activation of the unfolded protein response pathway without enhancing autophagocytic protein degradation [38, 42] . Consequently, inhibition of autophagy inhibits HCV replication [41] . Although most studies are carried out with HCV genotype 2a, it is now well accepted that autophagy probably also occurs with infection of other HCV genotypes. What is not clear from these studies is whether enhanced autophagy persists during the chronic phase of HCV infection, and if and to what extent oxidative stress contributes to HCV-mediated activation of autophagy. In this study, we observed the accumulation of autophagocytic vacuoles and up-regulation of the autophagosome marker, LC3, in the genome-length and subgenomic replicon cells (Figures 3B-3D) . The data suggests that HCV non-structural proteins play a role in inducing autophagy, possibly through ER stress ( Figure 1B) , mitochondrial damage ( Figures 1B and 2A) , and induction of oxidative stress (Figures 2D and 4B ). Core-on cells, on the other hand, had a slight increase in the number of autophagocytic vacuoles without a significant increase in LC3-II or total LC3 levels ( Figures 3B-3D) . HCV-mediated oxidative stress was likely a contributor in the activation of autophagy because enhanced antioxidant capacity significantly reduced total LC3 and LC3-II levels ( Figure 4A ), whereas increased oxidative stress led to a further increase in total LC3 and LC3-II ( Figure 4B ). ROS generated in the cytosol and mitochondria was equally capable of enhancing autophagy because increased antioxidant capacity in either compartment achieved comparable levels of reduction. In addition, H 2 O 2 appeared to be the main culprit in ROS-mediated activation of autophagy within this experimental system since addition of catalase effectively suppressed X/XO-mediated activation of autophagy ( Figure 4B ). Despite up-regulation of multiple antioxidant enzymes in subgenomic replicon cells ( Figure 5 ), these cells responded to further increase in antioxidant enzymes with reduced autophagy. It is important to note that although the antioxidant enzymes clearly can modulate the effects of viral proteins, they only serve to reduce the LC3 increase by ,26% in subgenomic replicon cells. The data suggest either that other mechanisms, such as ER stress, are important as well or that the antioxidant effect is incomplete. Overexpression of antioxidant enzymes also led to a reduction of autophagy in Core-on and Core-off cells. Antioxidant enzyme overexpression in cells with genome-length replicon, on the other hand, failed to reduce the level of autophagy even though the enhanced antioxidant capacity from SOD/CAT expression vectors was comparable to that of other cells ( Figure S4 ). In contrast, increased oxidative stress effectively increased autophagy in genome-length replicon cells ( Figure 4B ). The data suggest that oxidative stress may not play a significant role in the basal level of autophagy in genome-length replicon cells; however, the cells are sensitive to additional oxidative stress and are capable of accumulating more autophagosomes under conditions of increased oxidative stress. Although total LC3 levels were increased in X/XO-treated subgenomic replicon cells, LC3-II levels were not changed with the added oxidative stress ( Figure 4B ). It is possible that the conversion of LC3-I to LC3-II was already at the maximum level in subgenomic replicon cells, and additional oxidative stress was not able to enhance the reaction further. NS5A and, to a lesser extent, NS5B and the core protein have been detected in autophagocytic vacuoles (Table S1 ). Whether the NS5A localization in the autophagosome observed in our study is a cellular defense mechanism elicited to degrade foreign antigens or a viral mediated self-preserving mechanism to enable viral replication remains to be determined. HCV transgenic mice expressing the full-length HCV polyprotein were also examined for the accumulation of autophagocytic vacuoles in the liver. However, no evidence of enhanced autophagy was detected. It is possible that additional factors are needed to cause the accumulation of autophagosomes in vivo or that very low viral protein expression levels in the transgenic mice were not sufficient to induce these changes. A previous study using the same line of transgenic mice [43] showed that long-term iron overload was necessary to induce the accumulation of autophagosomes. Since iron overload increases oxidative stress in the liver, the result is consistent with our finding in HCV protein-expressing cells in the relationship between oxidative stress and autophagy. Huh7 cells with the subgenomic replicon showed significant upregulation of multiple antioxidant enzymes that belonged to the cytosolic and mitochondrial compartments ( Figure 5 ). Despite the up-regulation of multiple antioxidant enzymes, the ratios of oxidized (Sulfonic form) to total peroxiredoxin 1 and 3 in the subgenomic replicon cells remained the same as that in Huh7 cells ( Figure S5 ), suggesting that the overall redox environment was still more oxidizing. This conclusion was also supported by the increased MitoSOX staining and reduced aconitase activities ( Figures 2C and 2D and Figure S3B ). Whether the up-regulation of multiple antioxidant enzymes in the subgenomic replicon cells is a direct response to the presence of HCV non-structural proteins or is merely a clonal variation will need to be deciphered with additional studies in the future. However, it is worth noting that previous studies with different HCV subgenomic repliconexpressing cells also showed a similar up-regulation in various antioxidant enzymes [17, 44] . Huh7 cells with the genome-length replicon had a 1.8-fold increase in the mitochondrial form of thioredoxin (TRX2), but a 15% reduction of its upstream enzyme PRDX3 ( Figure 5 ). The net result was a slightly higher ratio of PRDX3-SO3 to total PRDX3; however, the increase was not significantly higher than that of Huh7 cells ( Figure S5B ). The expression level of non-structural proteins should be comparable between the genome-length and subgenomic replicon cells based on the SEAP reporter assay ( Figure S1A ) and published results [45] . Therefore, the lack of antioxidant response in cells with the genome-length replicon was probably not due to a difference in the expression of non-structural proteins; but rather, it could be due to the additional cytopathic effects from core, E1, E2, and p7 proteins that counteract the antioxidant enzyme response induced by non-structural proteins. Consequently, the genome-length replicon cells had a prolonged population doubling time and reduced cyclin D1 expression (Figure 6) , which suggested delayed cell cycle progression. In contrast, Core-on cells showed no changes in any of the antioxidant enzymes examined. In summary, our studies using Huh7 cells expressing different parts of the HCV polyprotein suggest that the presence of the fulllength HCV polyprotein leads to the most severe cellular damage, including generation of ROS, accumulation of lipids, ER stress, mitochondrial injury and degeneration, accumulation of autophagosomes, and prolonged population doubling time. Despite the elevated levels of ROS, these cells failed to mount a robust antioxidant response. Huh7 cells with only the non-structural proteins, on the other hand, have the same cytopathic changes as that observed in the genome-length replicon cells, but the cells showed a robust antioxidant response even though the response was not sufficient to suppress HCV-mediated ROS production. Our data suggest that HCV-mediated ER stress, mitochondrial injury, and oxidative stress form the three arms of mediators for the activation of autophagy (Figure 7) . Their effects are additive and inter-related. Without a concomitant increase in downstream protein degradation, autophagosomes accumulate and provide a sanctuary for HCV replication and protection from host immune surveillance [42] . The process likely helps to sustain a low level of viral production, which perpetuates chronic HCV infection and chronic liver injury (Figure 7) . In addition to activation of autophagy, HCV-mediated ER stress can lead to imbalance in Ca homeostasis, which further contributes to cellular oxidative stress [14, 15, 46, 47] . Mitochondrial injury can lead to metabolic deficits Figure 7 . Diagrammatic presentation of the cytopathic effects caused by various HCV proteins. Replication of HCV RNA and production of HCV structural (core, E1, E2, and p7) and non-structural proteins (NS2, NS3, NS4A, NS4B, NS5A, and NS5B) in the rough ER and the subsequent partition of HCV proteins to different subcellular compartments lead to ER stress, mitochondrial injury, and the production of ROS. These cytopathic effects lead to activation of autophagy without a concomitant increase in protein degradation. Consequently, autophagosomes accumulate in HCV infected cells. EM photo with normal mitochondria (M) and rough ER (RER) was taken from healthy Huh7 cells; EM photos with enlarged mitochondria, ER blebbing, lipid droplet (L), and autophagosomes (A) were taken from genome-length and subgenomic replicon cells. The HCV genetic materials contained in genome-length replicon, subgenomic replicon, and Core-on cells are depicted at the top of the diagram. doi:10.1371/journal.pone.0028551.g007 and further exacerbates cell injury and dysfunction. Oxidative stress can cause increased DNA damage and mutation, inactivation of redox-sensitive proteins, and activation of redox sensitive signaling pathways such as MAPK and AP-1 [48] [49] [50] . These cytopathic effects work in concert in the course of chronic HCV infection to promote cell death, hepatocyte turnover, alteration of the liver microenvironment, and ultimately cell transformation and tumorigenesis. All animal procedures were reviewed and approved under the protocol number HUT100602MOU by the VA IACUC committee (Subcommittee on Animal Studies, NIH assurance number A3088-01) at the VA Palo Alto Health Care System and in accordance with the PHS Policy on Humane Care and Use of Laboratory Animals. Cell culture and mouse model Control Huh7 cells and Huh7 cells with genome-length replicon, subgenomic replicon, or the tet-inducible core expression construct were used for this study. All three HCV proteinexpressing cell lines have been described previously [20, 26] . Genome-length replicon and subgenomic replicon were derived from a genotype 1a H77c infectious molecular clone [20] . The genome-length replicon encodes the full-length HCV polyprotein (core, E1, E2, p7, and NS2-NS5), whereas the subgenomic replicon encodes only the non-structural proteins NS2-NS5B [20] . The tet-inducible core-expressing cells express core protein in the absence of doxycycline. These cells are referred to as Core-on cells when core expression is turned on and as Core-off cells when core expression is turned off by the addition of doxycycline to the culture medium. All cells were maintained in DMEM with 10% tetracycline-free fetal calf serum (FCS, Clontech, Mountain View, CA, USA), Pen/Strep (50 IU/ml penicillin and 50 mg/ml streptomycin), and G418 (200 mg/ml) plus the following antibiotics: genome-length and subgenomic replicon cells, 5 mg/ml blasticidin; Core-off cells, 20 mg/ml doxycycline. When cells were plated for an experiment, only DMEM with 10% tetracycline-free FCS and Pen/Strep was used. All cells were incubated at 37uC with 5% CO 2 . Expression of HCV proteins in genome-length and subgenomic replicon cells were monitored by following the specific activities of the reporter gene, secreted alkaline phosphatase (SEAP), in the culture medium at 48 to 72 hrs after the initial plating [51] ; HCV core protein expression in genome-length replicon and Core-on cells were monitored by western blot analysis. HCV transgenic mice expressing the full-length HCV polyprotein [12] were used at 3 months of age. Only male mice were used in this study. All mice were kept in a barrier facility with a 12hr dark-light cycle, given food and water ad libitum, and maintained in microisolators with a constant temperature between 20uC and 26uC. All animal procedures were reviewed and approved by the IACUC committee at the VA Palo Alto Health Care System and in accordance with the PHS Policy on Humane Care and Use of Laboratory Animals. Control Huh7 cells and genome-length replicon, subgenomic replicon, Core-on, and Core-off cells were fixed with 2.5% glutaraldehyde in 0.1 M Na Cacodylate, pH 7.4 for 2 hrs, postfixed with 2% aqueous osmium tetroxide for 2.5 hrs, and subsequently stained en bloc in 2.5% uranyl acetate (in water) overnight before dehydration and embedding in Eponate 12 resin (Ted Pella, Inc., Redding, CA, USA). Thick (1 mm) sections of the embedded cells were examined at the light microscopic level. The cell blocks were further trimmed to obtain thin sections (80 nm), stained with saturated solution of uranyl acetate (15 min) followed by Reynolds' lead citrate (8 min) , and examined with a JEOL JEM 100CX II transmission electron microscope (JEOL Ltd., Tokyo, Japan). HCV transgenic and non-transgenic mice were perfused through the left ventricle of the heart, first with 10 U/ml heparin in saline until the liver was cleared of blood (about 5 min), then with fixative (2% glutaraldehyde and 2% paraformaldehyde in 0.1M Na Cacodylate, pH 7.4) until the liver was completely fixed (about 6-8 min). Heparin and fixative solutions were delivered by a peristaltic pump (VWR, Westchester, PA, USA) with the flow rate set at 4.5 ml/min. One mm cubes were prepared from the fixed liver and were left to continue fixation at room temperature (RT) overnight. These specimens were processed and embedded in Eponate 12 resin as described above. For mitochondrial size determination, electron micrographs taken at 7.2K magnification were used, and mitochondrial sizes from 3-4 cells from each cell line were measured. Image J was used to determine the area occupied as pixel number, which was then converted to mm 2 . For determination of the number of mitochondria and autophagocytic vacuoles/vesicles, electron micrographs taken at low magnifications (1.9-3.6K), with the requirement of being able to fit an entire cell into the view, were used and 10-26 cells from each cell line were analyzed. For live MitoSOX and MitoTracker staining and imaging, 2610 4 cells were seeded onto each chamber of tissue culture treated 8-chamber slides (Millicell EZ slides, Millipore, Billerica, MA, USA) and incubated overnight. For MitoSOX staining to detect superoxide generation in the mitochondria, culture media was replaced with 200 mL of staining solution containing 5 mM MitoSOX Red (Invitrogen, Carlsbad, CA, USA) and 5 mg Hoechst 33342 (for nuclear staining, Invitrogen) in HBSS, and cells were incubated at 37uC for 10 minutes. For equilibration, staining solution was replaced with 500 mL pre-warmed HBSS and returned to 37uC for another 10 minutes before imaging. For MitoTracker staining to visualize the mitochondrial network, culture media was replaced with 100 mL of staining solution containing 200 nM MitoTracker Green FM (Invitrogen) and 5 mg Hoechst 33342 in HBSS, and cells were incubated at 37uC for 30 minutes. The staining solution was then replaced with 500 mL pre-warmed HBSS before imaging. Cells were imaged directly in HBSS with a 40x (NA = 0.6) objective on an Olympus IX71 inverted fluorescence microscope equipped with a Coolsnap HQ monochrome camera (Photometrics, Tucson, AZ, USA). For direct comparison of staining intensity, exposure time was kept constant at 500 msec for MitoSOX and at 200 msec for MitoTracker. Staining intensities of MitoSOX and MitoTracker were determined with Image J as pixel intensities, and a minimum of 40 cells each were analyzed. Expression level of the reporter gene, secreted alkaline phosphatase (SEAP), was used to monitor the expression of the HCV genome-length and subgenomic replicons. Huh7 cells with the genome-length or subgenomic replicon were cultured to ,80% confluency and cell culture medium from each cell line was removed for SEAP assays. The SEAP reporter assay was performed using the Great EscAPe SEAP Fluorescence Detection Kit (Clontech), and SEAP activities were normalized to total cellular proteins. To determine total aconitase activities, Huh7 and HCV protein-expressing cells were cultured to 90% confluency in 60 mm plates. Cell pellets were resuspended in aconitase buffer (50 mM Tris-Cl, pH 8, 2 mM citric acid, and 0.6 mM MnCl 2 ) and passed through two rounds of freeze-thaw cycle between liquid nitrogen and room temperature water to break open membranes. Due to the low level of aconitases in Huh7 cells, total aconitase activities were determined with a kinetic assay [24] coupled to the PMS/MTT color reaction to enhance the sensitivity [25] . Following a 3-minute preincubation at 37uC in the dark, the reaction was continued at 37uC and the OD change was monitored every 5 minutes at 590 nm for up to 60 minutes using a plate reader (SpectraMax M3, Molecular Devices, Mountain View, CA, USA). The reaction appeared to be most linear in the first 30 minutes and consequently, OD change per minute, as a function of enzyme activities, was calculated for that time period. The final results were normalized to the amount of total proteins in the cell lysates. To enhance cellular antioxidant capacity, expression constructs for dual expression of CuZnSOD/cytosolic catalase (SOD1/ cCAT) or MnSOD/mitochondrial catalase (SOD2/mCAT) were used (constructed by SKZ, unpublished data). Expression of SOD1 and SOD2 are controlled by elongation factor-1 alpha (EF-1á) promoter, and expression of catalase is controlled by CMV promoter. Cells were grown to 50% confluency in 6-well plates and were transfected with 2.5 mg of each expression construct using TransITH-LT1 Transfection Reagent (Mirus Bio LLC, Madison, WI, USA). Cell lysates were prepared 36 hrs after the transfection for western blot analysis of LC3. CuZnSOD, MnSOD, and catalase levels were also determined to ensure comparable expression of each protein among different cell lines. To increase oxidative stress, cells were grown to 50% confluency in complete culture medium in 6-well plates and were then switched to 2% FCS-containing medium with 0.25 mM xanthine (X) and 20 mU/ml xanthine oxidase (XO). Since X/XO generates a combination of superoxide and H 2 O 2 , catalase (CAT, 40 mU/ml) was added to a subset of cultures to eliminate H 2 O 2 . Cells were incubated with X/XO or X/XO/CAT for 72 hrs; culture medium was changed every 24 hrs to maintain a steady level of X, XO, and CAT. Several immunochemical procedures are described below; antibodies used in this study are listed in Table 1 . Immunogold labeling. For immunogold EM localization of HCV proteins, cells were fixed (0.1% glutaraldehyde, 4% paraformaldehyde in phosphate buffered saline) and embedded in LR Gold resin (Ted Pella) using techniques described by Berryman and Rodewald [52] . Antibodies were pre-absorbed with Huh7 cell homogenate for 1-24 hrs. Thin sections on grids were blocked with 3% bovine serum albumin in TBS (Tris buffered saline) for 1 hr at RT, incubated with primary antibodies for 1 hr at RT, followed by gold-labeled goat anti-mouse IgG or goat antirabbit IgG for 1 hr at RT. The antibody complexes were stabilized with 2% glutaraldehyde in water, and the sections were stained with osmium vapor and lead citrate and examined as described above [53, 54] . Huh7 and Core-off cells were used as negative controls for antibody bindings. Multiple non-overlapping areas were scored for the frequency of HCV antigen localization and for the abundance of the antigen in each area. A positive signal is defined as the presence of at least two gold particles in a given organelle. Immunocytochemistry staining for LC3. To determine the status of LC3 protein, cells were seeded on 12 mm Fisherbrand Coverglass for Growth (Fisher Scientific, Pittsburgh, PA, USA) and cultured in a 24-well plate overnight. Huh7 cells treated overnight with Bafilomycin A1 (200 nM, Wako Chemicals, Richmond, VA, USA) were used as positive controls for autophagy. Cells were fixed in 4% paraformaldehyde (in PBS, pH 8), quenched in 50 mM NH 4 Cl for 10 min, permeabilized in 0.1% Triton X-100/PBS for 10 min, and blocked in 1% BSA for 10 min. Cells were then incubated with 15 ml rabbit anti-LC3 polyclonal antibody for 60 min, followed by goat anti-rabbit IgG conjugated with Alexa Fluor 488 for 30 min. Cell nuclei were stained with 10 mg/ml Hoechst 33258 (Invitrogen). All steps were carried out at RT. Cover slips were washed in water, mounted in a drop of ProLong Gold antifade reagent (Invitrogen), and air-dried in the dark overnight. Cells were imaged with a 40x (NA = 0.75) objective on a Zeiss AxioVision microscope equipped with a Hamamatsu monochrome camera. For direct comparison of staining intensity, exposure time was kept constant at 89 msec. Western blot analyses. To determine the protein level of HCV core protein, CuZn superoxide dismutase (CuZnSOD), Mn superoxide dismutase (MnSOD), Peroxiredoxin 1 and 3 (PRDX1 and 3), sulfonylated peroxiredoxin (PRDX-SO3), and cytosolic and mitochondrial aconitase (ACO1 and ACO2), cell pellets were incubated on ice with Tissue Protein Extraction reagent (T-PER, Fisher Scientific) supplemented with a protease inhibitor cocktail (Roche, Indianapolis, IN, USA), passed through 26 gauge needles several times, and centrifuged at 13,000 g, 4uC, for 2 min to remove cell debris. Total protein concentration was determined with a NanoVue Spectrophotometer (GE Healthcare, Piscataway, NJ, USA), and 50 ìg total proteins per lane were used for western blot analyses. Proteins were separated by 4-20% Mini-PROTEANH TGX gels (Bio-Rad, Hercules, CA, USA) (core, CuZnSOD, MnSOD, ACO1, and ACO2) or 16% Novex Tris-Glycine gels (Invitrogen) (PRDX1, PRDX3, and PRDX-SO3) and transferred to 0.2 mm nitrocellulose membranes (Bio-Rad). PRDX-SO3 antibody only detects sulfonylated PRDX and is not specific to PRDX1-SO3 or PRDX3-SO3. Consequently, sulfonylated PRDX1 and PRDX3 were distinguished based on their size difference ( Figure S5C ). To determine LC3 protein levels, cells cultured in 60 mm dishes were washed once with PBS and then scraped directly in the cell lysis buffer (50 mM Tris-Cl, pH 7.6, 150 mM NaCl, 2% SDS, 1 mM EDTA, 1x protease inhibitor cocktail) on ice. To break open the membranes, cell lysates were freeze-thawed two times and passed through a 26 gauge needle five times. Fifty mg of each cell lysate was separated with Any KD Mini-PROTEANH TGX gels (Bio-Rad) and transferred to 0.2 mm nitrocellulose membranes. The same membranes were used for thioredoxin 1 (TRX1) and thioredoxin 2 (TRX2) analyses. Alternatively, LC3-I and II were separated by 12% NuPAGE Bis-Tris gels (Invitrogen) in a subset of the studies. All membranes were incubated with primary antibodies overnight at 4uC followed by IR-labeled secondary antibodies for 1 hr at RT on a shaker. Membranes were washed three times each between primary and secondary antibody incubation and after secondary antibodies with PBST (PBS with 0.1% Tween 20). Membranes were then analyzed with the Odyssey Infrared Imaging System (Licor Biosciences, Lincoln, NE, USA). Signal intensity of each protein was normalized to that of âactin. Three to four independent experiments were carried out for each protein; all results are plotted as mean 6 SEM. The statistical analysis program GraphPad Prism (version 4.03, GraphPad Software, Inc., San Diego, CA, USA) was used for data analyses. One-way ANOVA analyses with Dunnett's post-hoc test were carried out initially to see if there was an effect of HCV proteins or treatments on the cellular indices being measured. Two-tailed Student's t tests were then used for pair-wise comparison between Huh7 and Genome-length replicon, subgenomic replicon, or Core-on cells. Student's t tests were also carried out for comparisons between Core-on and Core-off cells. Figure S1 Genome-length replicon, subgenomic replicon, and core protein expression. A, SEAP activities are used to monitor the expression of genome-length and subgenomic replicons in Huh7 cells. B, western blot analysis is used to determine the expression of core protein in genome-length replicon cells and Core-on cells. (TIF) Figure S2 Oil-red-O staining for lipid deposit in HCV proteinexpressing Huh7 cells. Nuclei are stained with hematoxylin. Pictures were taken using a 20x objective. (TIF) Figure S3 Cytosolic aconitase (ACO1) protein levels and total aconitase activities. ACO1 protein levels (A) were determined by western blot analyses, and total aconitase activities (B) were determined with a kinetic assay coupled to the PMS/MTT color reaction. Mitochondrial aconitase (ACO2) protein levels were too low to be reliably quantified. Data are presented as mean 6 SEM of three independent experiments. (TIF) Figure S4 Overexpression of superoxide dismutase (SOD) and catalase (CAT) in HCV protein-expressing Huh7 cells. Cells were transfected with expression vectors designed for dual expression of CuZnSOD/cytosolic catalase (SOD1/cCAT) or MnSOD/mitochondrial catalase (SOD2/mCAT) to increase antioxidant capacity in the cytosol or mitochondria, respectively. For each set of transfection, the order of cells loaded (from left to right) is Huh7, genome-length, subgenomic, Core-on, and Core-off cells. SOD1 and SOD2 are tagged with V5 and CAT with Myc epitope and are detected with antibodies against these epitopes. Endogenous SOD1, SOD2, and CAT are not detectible by V5 or Myc antibody and are therefore, not visible in this blot. (TIF) Figure S5 The redox state of peroxiredoxin 1 (PRDX1) and 3 (PRDX3). The ratios of PRDX1-SO3 to PRDX1 (A) and PRDX3-SO3 to PRDX3 (B) were determined by sequential western blot analyses (see C below). C, representative western blots showing the separation of PRDX1 and PRDX3 in 16% polyacrylamide gels and the sequential binding of specific antibodies to PRDX-SO3, PRDX1, and PRDX3. Data are presented as mean 6 SEM of four independent experiments. (TIF) Table S1 Immunogold EM localization of HCV proteins. The subcellular location of HCV core, NS5A, and NS5B proteins and the frequency at which they are observed in each organelle by immunogold electron microscopy is presented. (DOCX)

Neighborhood Properties Are Important Determinants of Temperature Sensitive Mutations

Temperature-sensitive (TS) mutants are powerful tools to study gene function in vivo. These mutants exhibit wild-type activity at permissive temperatures and reduced activity at restrictive temperatures. Although random mutagenesis can be used to generate TS mutants, the procedure is laborious and unfeasible in multicellular organisms. Further, the underlying molecular mechanisms of the TS phenotype are poorly understood. To elucidate TS mechanisms, we used a machine learning method–logistic regression–to investigate a large number of sequence and structure features. We developed and tested 133 features, describing properties of either the mutation site or the mutation site neighborhood. We defined three types of neighborhood using sequence distance, Euclidean distance, and topological distance. We discovered that neighborhood features outperformed mutation site features in predicting TS mutations. The most predictive features suggest that TS mutations tend to occur at buried and rigid residues, and are located at conserved protein domains. The environment of a buried residue often determines the overall structural stability of a protein, thus may lead to reversible activity change upon temperature switch. We developed TS prediction models based on logistic regression and the Lasso regularized procedure. Through a ten-fold cross-validation, we obtained the area under the curve of 0.91 for the model using both sequence and structure features. Testing on independent datasets suggested that the model predicted TS mutations with a 50% precision. In summary, our study elucidated the molecular basis of TS mutants and suggested the importance of neighborhood properties in determining TS mutations. We further developed models to predict TS mutations derived from single amino acid substitutions. In this way, TS mutants can be efficiently obtained through experimentally introducing the predicted mutations.

0.38 -0.03 0.48 1.11 0.00 Turn breaker: turn is defined by DSSP [2] 0.58 -0.03 0.48 0.32 0.00 * ACC = accuracy, MCC = Matthews correlation coefficient, AUC = area under the curve, KL = Kullback-Leibler divergence, DD = distribution distance. These values were calculated from a ten-fold cross-validation of each feature.

Missing and accounted for: gaps and areas of wealth in the public health review literature

BACKGROUND: High-quality review evidence is useful for informing and influencing public health policy and practice decisions. However, certain topic areas lack representation in terms of the quantity and quality of review literature available. The objectives of this paper are to identify the quantity, as well as quality, of review-level evidence available on the effectiveness of public health interventions for public health decision makers. METHODS: Searches conducted on http://www.health-evidence.ca produced an inventory of public health review literature in 21 topic areas. Gaps and areas of wealth in the review literature, as well as the proportion of reviews rated methodologically strong, moderate, or weak were identified. The top 10 topic areas of interest for registered users and visitors of http://www.health-evidence.ca were extracted from user profile data and Google Analytics. RESULTS: Registered users' top three interests included: 1) healthy communities, 2) chronic diseases, and 3) nutrition. The top three preferences for visitors included: 1) chronic diseases, 2) physical activity, and 3) addiction/substance use. All of the topic areas with many (301+) available reviews were of interest to registered users and/or visitors (mental health, physical activity, addiction/substance use, adolescent health, child health, nutrition, adult health, and chronic diseases). Conversely, the majority of registered users and/or visitors did not have preference for topic areas with few (≤ 150) available reviews (food safety and inspection, dental health, environmental health) with the exception of social determinants of health and healthy communities. Across registered users' and visitors' topic areas of preference, 80.2% of the reviews were of well-done methodological quality, with 43.5% of reviews having a strong quality rating and 36.7% a moderate review quality rating. CONCLUSIONS: In topic areas in which many reviews are available, higher level syntheses are needed to guide policy and practice. For other topic areas with few reviews, it is necessary to determine whether primary study evidence exists, or is needed, so that reviews can be conducted in the future. Considering that less than half of the reviews available on http://www.health-evidence.ca are of strong methodological quality, the quality of the review-level evidence needs to improve across the range of public health topic areas.

Using Systematic Reviews A systematic review consists of an examination of all of the primary studies on a topic, which includes searching for, collating, and assessing the studies, to establish conclusive evidence about a topic [1] . The Cochrane Collaboration is an international body that produces systematic reviews of primary research at the highest standard, and as such, this is a commonly accepted definition of systematic reviews. Evidence-informed public health advocates the incorporation of the best available scientific evidence into decision making [2] . Review level evidence is an important part of evidence-informed public health decision making, since reviews synthesize the results of individual studies, providing a more accurate estimate of the effects of an intervention [3] . Rigorous synthesis of primary research minimizes bias [4] [5] [6] , explains differences among studies relating to the same research question [7, 8] , and presents more precise and consistent summary statistics than the effect sizes found in individual studies [5, [8] [9] [10] . Well-conducted reviews provide high-quality, accurate evidence [4] [5] [6] , increasing decision-makers' confidence in the strength of the review evidence and in applying the findings in practice [4] . Public health decision makers prefer using systematic reviews to assist in decision-making given that review level evidence saves time and is more efficient compared to using primary studies [6, 11] . Systematic review findings can be generalized to a larger sample, providing a great evidence base for users; such external validity is essential to ensure adaptability and applicability of evidence-based interventions into the local context [12] . Consequently, systematic reviews are useful for informing and influencing public health policy and practice decisions [7, 13, 14] . While the value of review level evidence is acknowledged and well documented, public health decision makers encounter a number of challenges in incorporating systematic reviews in their decision making. Review level evidence is available in journals and obtainable through bibliographic database searches [15] , yet barriers still exist in accessing the information. Public health decision-makers often have difficulties locating systematic reviews in the published literature due to database indexing limitations, limited availability of relevant public health reviews, lack of primary study evidence and thus a lack of reviews in some topic areas, and a lack of interest in certain topic areas by researchers conducting reviews [3, 15, 16] . Even when systematic reviews are identified, only a small proportion of those are relevant to public health. For example, one search strategy captured 41, 871 abstract titles across all research topics in public health but once screened, only 1, 356 were identified as being potentially relevant, of which only 207 reviews were actually deemed relevant to public health [16] . The majority of published systematic reviews pertain to clinical topics rather than public health [16] . Consequently, there are gaps at the systematic review level across the spectrum of public health practice [6, 17] . When relevant reviews are located, users still need to be critical of that evidence. There are a number of search engines that provide evidence from various databases, such as the Cochrane Database of Systematic Reviews, PubMed, and the Campbell Collaboration, but the evidence is not critically appraised [17, 18] . While some of these databases which include public health relevant evidence assess the quality of the evidence, many do not [19] . To reduce bias in evidence-informed practice, public health decision makers need to be able to assess the methodological quality of systematic reviews [4] . However, critical appraisal skills have been identified as a significant barrier to using research evidence in decision making [20] . Development of individual capacity is important in addressing appraisal challenges as well as providing support [21] . Health Evidence is a research and service organization aimed at supporting Canada's public health decision makers in accessing and interpreting research evidence. The target audience for Health Evidence includes medical officers of health, policy makers, program managers, and frontline workers in public health. Given the audience, decision making may take place at the local level (such as public health units/regional health authorities), provincial level (such as ministries), or federal level (such as government). Our most widely accessible resource is the http://www.health-evidence.ca online registry of systematic reviews; a free, user-friendly, searchable database of public health relevant, qualityappraised systematic reviews published since 1985 evaluating the effectiveness of public health interventions. Given that unpublished literature, such as conference abstracts, provide little added value [22] , Health Evidence mainly focuses on published review literature. In order to identify the scope of interventions to include in the health-evidence.ca registry, qualitative interviews were conducted, as well as seeking organizational charts and information from every province and territory in Canada on the services public health units provide. Systematic reviews are considered relevant if: 1) the article is a review, which includes the synthesis of more than one primary study; 2) the intervention is relevant to public health practice; 3) the effectiveness of an intervention is evaluated; 4) the evidence on health outcomes is reported; and 5) the search strategy is described [19] . To assess the methodological quality, the following ten criteria are used: 1) a clearly focused question was stated; 2) inclusion criteria were explicitly stated; 3) a comprehensive search strategy was described; 4) an adequate number of years were covered in the search; 5) a description of the level of evidence was provided; 6) the methodological rigor of primary studies was conducted and results were described; 7) the methodological quality of primary studies was assessed by two reviewers and the level of agreement was provided; 8) tests of homogeneity or assessment of similarity of results across studies was conducted and reported; 9) appropriate weighting of primary studies was conducted; and 10) the author's interpretation of the results were supported by the data [19] . Each criterion is equally weighted and a final methodological score is tallied out of 10. Reviews with an overall rating of eight or more are considered strong, five to seven, moderate, and below four are considered to be weak in methodological quality. Due to competing demands, it is necessary for decision makers to quickly find, assess and use evidence to inform their decision making. The health-evidence.ca registry eliminates the need for users to search individual databases, identify relevant reviews, and conduct critical appraisal on the effectiveness of public health interventions. The tools used by Health Evidence to assess relevance and conduct critical appraisal are available online http://health-evidence.ca/html/HowJudgefor-Yourself, accessed 6 May 2011), and users can view completed critical appraisal tools for each review in the registry. In order to reach public health decision makers, Health Evidence is promoted at conferences, workshops, and site-visits, through outreach and engagement, networking, and listservs via website posts and e-newsletters, and through social media, such as Twitter and YouTube. Health Evidence also connects with public health decision makers through various partnerships and collaborations with the National Collaboration Centres for Public Health, public health units, the Canadian Best Practices Portal, and the Public Health Agency of Canada. The registry is also listed as a resource on several public health organization and university websites, such as Research into Action, Pan American Health Organization, KT+ Knowledge Translation, Canadian Institute for Health Research (CIHR) Knowledge Translation and Commercialization, Nova Southeastern University, Dalhousie University, and more. Health-evidence.ca has nearly 5, 000 registered users, and sees over 40, 000 visitors annually representing more than 150 countries. In the development of the health-evidence.ca registry, 21 topic areas of interest to public health decision makers were identified through focus groups and consultations with key informants within the public health setting. The purpose of the registry is to facilitate access to review-level evidence for decision makers working in program planning and policymaking in public health and health promotion [19] . All reviews in the registry are indexed according to these 21 Focus of Review topic areas allowing site visitors to search the registry using common public health terms. In addition, each registered user completes a profile when signing up to the site checking off as many of the 21 topic areas relevant to them. This enables each registered user to receive a list of reviews related to their areas of interest, along with a rating of the methodological quality of each review, each quarter when the registry is updated. Unfortunately, for some public health topics, there are limited or no high quality reviews available and for others the reviews that are available are not of good methodological quality, meaning that use of these findings in decision making requires careful consideration. A thorough search of http://www.health-evidence.ca allowed us to indentify the top areas of interest to public health decision makers, and provide an overview of the availability of review-level evidence within these areas. In this paper we will not only identify topic areas of high interest to public health decision makers, we will also highlight existing gaps as well as identify topic areas with an abundance of high-quality evidence. One objective of this paper is to identify the quantity of systematic reviews available on the effectiveness of public health interventions, so as to encourage researchers and research funders to conduct/fund systematic reviews where gaps exist. A second objective is to identify the quality of systematic review evidence on the effectiveness of public health interventions in order to encourage higher quality methodological reviews and higher level synthesis of topics areas rich in high-quality reviews (e. g., review of reviews). Populating the health-evidence.ca registry of systematic reviews The health-evidence.ca registry of systematic reviews is populated through an extensive ongoing search (1985present) of seven electronic databases (MEDLINE, EMBASE, CINAHL, PsycINFO, Sociological Abstracts, BIOSIS, SportDiscus), handsearching of 46 journals, and screening the reference lists of all relevant reviews [19] . Reviews are assessed for relevance, and then relevant reviews are indexed by commonly-used public health terms and quality assessed by two independent reviewers who come to agreement on the final rating of each review (strong, moderate, weak). More detail on http://www.health-evidence.ca has previously been published [19] . Assessing health-evidence.ca user and visitor areas of interest Registered user areas of interest were assessed by querying the health-evidence.ca registered user database and looking at the areas of interest identified by all users who registered up to December 31, 2010. Data were aggregated by topic area. Registered user data is provided voluntarily by users and aggregation ensures individual data remain anonymous. Topic areas of interest were ranked from highest to lowest rates of user interest. The top 10 areas of interest were summed to generate the denominator: total user interest in the top 10 topic areas. Visitor areas of interest were assessed by summing frequency of visitor searches of the 21 Focus of Review topic areas and visitor use of the topic area browse menu for the period January 1, 2010 to December 31, 2010. Visitor site usage is tracked via Google Analytics, a web analytics tool that collects and aggregates nonpersonal data to report on visitor interaction with the health-evidence.ca website. Total search and browse access by unique visitors were ranked from highest to lowest pageviews. The top 10 areas of interest were summed to generate the denominator: total visitor interest in the top 10 topic areas. The health-evidence.ca registry was used to identify gaps and areas of wealth in the public health review literature. Each of the 21 Focus of Review topic areas were searched, and the quantity and proportion of reviews rated methodologically strong, moderate, and weak were identified. Three categories were used to define availability of reviews within each topic area: (+) few, representing 1-150 reviews; (++) moderate, representing 151-300 reviews; and, (+++) many, representing topic areas possessing greater than 301 reviews. Reviews that addressed multiple topics were accounted for within each topic area that they addressed (e.g., a review on the effectiveness of exercise in preventing chronic disease would be categorized as both physical activity and chronic disease). As of December 31, 2010, there were 4, 842 health-evidence.ca registered users, with each user identifying an average of 6.3 areas of interest, resulting in a total of 30, 363 identified topic areas. Upon registration, each user is asked to indicate as many areas of interest as they find relevant, which results in more identified areas of interest than total users. For the purpose of accurately representing the data showing all interest, we have included all indications in interest in each topic area, knowing that the denominator used represents total expressions of interest as opposed to total users. The top 10 registered users' topic areas are represented in Figure 1 . Registered users' top three topic areas out of the top 10 include by order of interest: 1) healthy communities, 2) chronic diseases, and 3) nutrition. order of interest include: 1) chronic diseases, 2) physical activity, and 3) addiction/substance use. The top 10 topic areas of interest of registered users and the top 10 topic areas of interest of visitors of http:// www.health-evidence.ca, as well as the availability of review evidence by methodological quality, are identified in Table 1 . The top areas of interest and the total number of reviews available included: addiction/substance use (355), adolescent health (367), adult health (552), child health (409), chronic diseases (702), communicable disease/infection (241), healthy communities (134), injury prevention/safety (296), mental health (336), nutrition (426), parenting (287), physical activity (353), reproductive health (240), and social determinants of health (66). While there was overlap between six of the registered users' and visitors' top areas of interest, the topic areas healthy communities, adult health, adolescent health, and, communicable disease/infection were preferred by registered users alone, and visitors had preferences for addiction/substance use, parenting, injury prevention/ safety, and, reproductive health. For the six areas of interest similar to both registered users and visitors, differences exist in the order of expressed interest. Three topic areas ranked similarly among the top 10 for both registered users and visitors with a difference of only one rank apart: chronic disease ranked high on both top 10 lists, ranking first for visitors and second for registered users; nutrition ranked third for registered users and fourth for visitors; and social determinants of health ranked sixth for visitors and seventh for registered users. The remaining three topic areas that both visitors and registered users were interested in included: mental health, ranking seventh for visitors and ninth for registered users; physical activity, ranking second for visitors and fifth for registered users, and, child health, ranking fourth for registered users and ninth for visitors. In the top five highest ranking common topic areas, both groups had interest in chronic diseases, nutrition, and, physical activity Characteristics of registered users of and visitors to the Health Evidence registry are provided in Table 2 . Based on the sample, 82.7% of registered users and 65.0% of visitors to health-evidence.ca are Canadian, and English is the language of preference for 98.1% of registered users and 83.3% of visitors. Upon registering to health-evidence.ca, users are asked to provide their organizational affiliation; 54.6% of users work in the field of public health or health services. As of December 31, 2010, 94.8% of registered users were subscribers to the quarterly Health Evidence tailored e-newsletter. As of April 1, 2011 there were 2, 175 systematic reviews evaluating the effectiveness of public health and health promotion interventions indexed in the health-evidence. ca registry. Table 3 provides an overview of the availability of reviews within each of the 21 Focus of Review topic areas. Figure 3 depicts the relationship between registered users' interests, visitor searches, and available reviews within each of the 21 topic areas. Topic areas with fewer than 150 reviews included: food safety and inspection (13) , dental health (62), social determinants of health (66), environmental health (69), and healthy communities (134). Of the areas with fewer than 150 reviews, healthy communities ranked first as an area of interest for registered users with 2, 548 registered users indicating interest in this topic. Social determinants of health ranked sixth and seventh for visitors * Visitor data for organization type is shown for organizations with five or more visits in 2010. † Visitor data was collected by internet service provider and many organizations could only be identified by internet service provider (e.g. Rogers, Shaw, Bell, etc.) and registered users respectively, with 1, 528 visitor searches submitted in 2010 and 1, 734 registered users indicating interest in the topic. The majority of healthevidence.ca registered users did not have preference for the other three identified areas with fewer than 150 reviews. Topic areas with a moderate number of reviews (151-300 reviews) included: infant health (153), senior health (152), sexual health (195), sexually transmitted infections (208), communicable disease/infection (229), reproductive health (232), parenting (282), and, injury prevention/safety (241). Four of these eight topic areas have expressed interest by visitors or registered users. For visitors, ranking at fifth, eighth and tenth place respectively, searches submitted in 2010 for parenting totalled 1, 813 pageviews, for injury prevention/safety 1, 360, and, for reproductive health 1, 273 pageviews. Ranking tenth place, 1, 422 registered users indicated interest in the topic communicable disease/infection. Topic areas with a large quantity of systematic reviews (301 or more reviews) include: mental health (336), physical activity (353), addiction/substance use (355), adolescent health (367), child health (409), nutrition (426), adult health (552), and chronic diseases (702). All of the topic areas with many available reviews (301+) were of interest to registered users and/or visitors. Chronic diseases, nutrition, and, physical activity were the three highest-ranking areas of interest common across both groups. The number one visitor area of interest and number two registered user area of interest was chronic diseases with 3, 115 visitor searches submitted in 2010, and 2, 153 registered users expressing interest in the topic. Nutrition ranked third for registered users and fourth for visitors with 1, 877 user interests, and 2, 010 visitor searches respectively. Physical activity ranked second for visitors with 2, 350 searches, and fifth for registered users with 1, 824 interested in the topic. Child health and mental health were two additional topics with many available reviews that rank in the top ten for both registered users and visitors. In the area of child health, 1, 846 registered users expressed interest and 1, 293 visitors submitted searches, ranking it fourth and ninth respectively, and the area of mental health ranked seventh for visitors with 1, 502 searches submitted, and ninth for registered users with 1, 459 interested in the topic. The remaining three topic areas with many available reviews were preferred by either visitors or registered users (i.e., not common to both groups). Addiction/substance use ranked third for visitors with 2, 283 search page views in 2010. Adult health and adolescent health ranked sixth and eighth respectively for registered users with 1, 778 and 1, 636 registered users indicating interest in each of these topics. A master list categorizing all of the 21 topic areas and the corresponding number of reviews available on the effectiveness of public health interventions, as well as the methodological quality, are listed in the additional file 1. The 21 Focus of Review topic areas were further broken down into 291 sub-topic categories. There were 34 sub-topics with no reviews available including: hormone replacement therapy, infertility, Norwalk virus, autism, and elder abuse, among others ( Table 4 ). The 21 Focus of Review topic areas that had sub-topics with no review included adult health, communicable disease/infection, dental health, environmental health, food safety and inspection, parenting, and senior health. The largest proportion of sub-topic with no review was observed within communicable disease/infection (n = 12). Adult health was ranked sixth and communicable disease/ infection ranked tenth by registered users. Parenting was ranked as a fifth preference for visitors. The remaining sub-topic with no reviews within dental health, environmental health, food safety and inspection, and senior health were not a preference for either registered users or visitors. (+)few reviews as 1-150, (++)moderate reviews as 151-300, (+++)many reviews as 301 and greater In addition there were 68 sub-topics with fewer than five reviews available, such as lung cancer, testicular cancer, food service inspection, fetal alcohol syndrome, sexual assault, and social justice. The full list of subtopics with fewer than five reviews is included in Table 5 . Most of the sub-topics with fewer than five reviews were within the registered users and visitors' topic areas of interest. Topic areas which were of interest to both registered user and visitors only had a small proportion of sub-topics with less than five reviews available (child health, chronic diseases, mental health, nutrition, and social determinants of health). Whereas the communicable disease/infection topic area, which ranked tenth among registered users, had the largest proportion of sub-topics with fewer than five reviews (n = 16). While environmental health and food safety and inspection were not preferred topic areas for registered users and visitors, a large portion of the subtopics in these two categories had fewer than 5 reviews. Although there is currently a lack of review literature in these areas, these topics have been mentioned by public health professionals as relevant to public-health practice. Also, there were numerous sub-topics with a great number of reviews available. Table 6 identifies 71 subtopics with more than 25 reviews available. Such subtopics included, but were not limited to, alcohol abuse/ use, smoking cessation, women's health, cancer, cardiovascular disease, lifestyle behaviours, disease transmission, depression, diet, healthy weight, exercise, and HIV. As well, most of these sub-topics were within the registered users and visitors' topic areas of interest, with the exception of dental health, senior health, sexual health, and sexually transmitted infections. The Health Evidence methodological quality rating is based on the ten criteria used to assess the strength of the methods. The proportion of reviews rated as having strong, moderate, or weak methodological quality was constant across all topic areas. 80.2% of the reviews on health-evidence.ca were of strong (43.5%) or moderate (36.7%) methodological quality. These well-done reviews included slightly more strong review quality ratings compared to the moderate review ratings. The remaining 19.8% of reviews in the top areas of interest were of weak methodological quality. These weak quality reviews met four or fewer methodological quality criteria, and as such, scored poorly on six or more of the ten criteria. Based on the weak reviews, 10.5% did not have a clearly focused question, 46.5% did not use appropriate inclusion criteria, 87.7% did not have a comprehensive search strategy, 34% did not cover an adequate number of years, 48% did not describe the level of evidence in the primary studies, 96.6% did not assess the methodological A number of sub-topic areas within public health featured no reviews or a very small number of reviews including those within the main topic areas for women's health (sub-topics include female genital mutilation, hormone replacement therapy, and infertility as examples); communicable disease/infection (sub-topics include food-borne diseases, hantavirus, and Norwalk, among others); food safety and inspection (sub-topics include botulism, food processing and inspection, and Hepatitis A as examples); dental health (sub-topic: dental implants); environmental health (sub-topics include asbestos, carbon monoxide poisoning, environmental epidemiology, extreme temperature, swimming pools, and flood, as examples). In these areas, review literature is needed to add to the existing body of literature from which decision makers can draw. In topic areas lacking in review literature, realist reviews, which provide explanatory analyses [23] , are a relatively new type of review that may provide further insight into the topic area. While these areas show the deficits in review literature, in 25 other sub-topic areas there was a wealth of systematic review literature of moderate or strong quality, including but not limited to, alcohol abuse/use, smoking cessation, women's health, cancer, cardiovascular disease, lifestyle behaviours, disease transmission, depression, diet, healthy weight, exercise, and HIV. In these areas offering many reviews, most had 15 or more which were of strong methodological quality. While review groups can identify and fill gaps in areas where evidence is lacking, there is also an opportunity to produce higher level syntheses where good-quality review evidence is available. Based on this analysis of the published, public health review literature catalogued in http://www.health-evidence.ca, all of the topic areas having many available reviews were also preferred areas of interest for users of the site. It is unclear how the availability of evidence is linked to interest in a topic, but it may be that demand can generate reviews in a particular topic area and that the public spurs research to fill gaps [24] . Alternately, preference for a topic may be a reflection of there being available evidence that drives interest in the topic and web site updates regarding that topic. Interestingly, the 15 priority topic areas indicated by a 2005 Cochrane global priority setting exercise [24] still closely match the top 10 topic areas of interest indicated by health-evidence.ca users and visitors. While resources such as http://www.health-evidence. ca, provide synthesized, high quality research evidence relevant to public health practice, coverage of health topics is not equal [25] . Compared to clinical review literature, there is far less population health systematic review literature [6] and within public health, a number of topic areas are without a solid base of review evidence evaluating the effectiveness of interventions (e.g., environmental health, social determinants of health) [18, 26, 27] . Public health studies are difficult to design and results are drawn from natural experiments (e.g. one health unit adopting a program compared to another health unit), thus fewer studies have been developed on the effectiveness of public health interventions compared to randomized controlled trials on medical treatments [28, 29] . However this lack of review-level evidence doesn't necessarily indicate a lack of evidence [30] . There may be primary research or other forms of evidence that can inform decisions but which may not yet have been synthesized. In cases where there are no eligible studies available to be reviewed for a particular topic, the result can be an "empty review", meaning that no studies were located which met the inclusion criteria to answer the question for review. Empty reviews can go unreported but may in fact be useful since empty reviews indicate interest in an area, highlight gaps, and offer a snapshot of the state of research evidence at the time of publication [31] . Even in light of a lack of evidence, or poor reporting of the evidence, decision makers can and should take an informed approach to having insufficient evidence [30] . An informed approach may be needed in topic areas that demonstrated a lack of review-level evidence, such as dental health, environmental health, food safety and inspection, and seniors' health, and particularly for public health priority areas such as healthy communities and social determinants of health. Consequently, the best literature may be sparse or of low quality in these particular topic areas. In these areas, new systematic review literature is needed to inform practice and policy decision making. It is unclear at this time whether gaps pertain only to reviews lacking in these topic areas, or whether there is a corresponding lack of primary studies as well, hindering the production of reviews. In some instances it may be necessary to first develop the primary study base in order for studies to be available for synthesis in a systematic review. Future reviews should be conducted on these broad topic areas for which review-level evidence on the effectiveness of interventions is lacking. Funding organizations should generate calls for syntheses to address these gap areas, while non-government and public health organizations should provide feedback on the lack of evidence in areas of interest to them to potential funders. Funding priorities for syntheses should reflect those areas which are priorities for public health decision makers both in Canada and internationally. There are promising indicators of demand for reviews, including actions being taken to promote the use of reviews [6, 11, 19, 30, [32] [33] [34] [35] [36] , an awareness of sites providing access to review-level evidence [19] and an increasing number of groups generating summaries of reviews [33] . Despite this heightened activity, given the gaps, a greater investment is needed to provide an evidence base that can meet demand and determine how to apply existing good quality systematic reviews in different contexts [25] . Organizations involved in the conduct of systematic reviews should direct synthesis funding to areas lacking in review content, or should consider higher-level reviews of reviews (where appropriate), where large bodies of review evidence already exist. Examples of organizations that conduct systematic reviews and reviews of reviews include: The Cochrane Database of Systematic Reviews http://www.cochrane.org/reviews/ index.htm, The Campbell Collaboration http://www. campbellcollaboration.org, The Centre for Reviews and Dissemination (CRD) http://www.york.ac.uk/inst/crd/ index_databases.htm, Health Technology Assessment international (HTAi) http://www.htai.org/, Effective Public Health Practice Project (EPHPP) http://www.ephpp. ca, CDC Guide to Community Preventative Services http://www.thecommunityguide.org, Canadian Agency for Drugs and Technology in Health (CADTH) http:// www.cadth.ca/, Agency of Healthcare Research and Quality (AHRQ) http://www.ahrq.gov/ and the National Institute for Health and Clinical Excellence (NICE) http://www.nice.org.uk. Some groups have a prioritization process to identify and meet needs for systematic reviews; for example, the Agency of Healthcare Research and Quality has a topic prioritization group to determine the relative importance of their effectiveness reviews against a standard set of criteria to prioritize unmet needs [37] . The Cochrane Collaboration aims to increase the quantity and quality of public health systematic reviews specifically [32] , with the new Health Promotion and Public Health Review Group announced in 2006 to prioritize and produce public health relevant reviews [38] . Authors have suggested that a global registry of anticipated public health studies could help to fill the gaps by making it easier to identify relevant, but potentially unpublished, primary studies available for review [6] . While well-done reviews in a large number of areas are available, it is important to continue to improve the quality of the overall body of public health review literature. Considering that the majority of weak reviews scored poorly on assessing the methodological quality of the primary studies, transparency, methods for combining or comparing results, conducting a comprehensive search strategy, and data supporting the author's interpretations, review authors should be cognisant of these criteria when conducting systematic reviews. In improving the quality of systematic reviews, the overall goals should ensure there are high quality reviews in all public health topic areas and a higher standard across the board. Although this paper provides information on the quantity and quality of systematic reviews on the effectiveness of public health interventions, there are a few limitations, including the lack of information regarding visitors to the Health Evidence registry. While the majority of visitors access the registry from various internet service providers, we are unable to determine their organization and the relevance of the registry content to a particular organization. Additionally, it is unclear at this time whether visitors and registered users are using reviews housed in the registry to inform their practice decisions. Studies are ongoing to evaluate the registry's usefulness and effectiveness.

Inhibition of Interferon Induction and Action by the Nairovirus Nairobi Sheep Disease Virus/Ganjam Virus

The Nairoviruses are an important group of tick-borne viruses that includes pathogens of man (Crimean Congo hemorrhagic fever virus) and livestock animals (Dugbe virus, Nairobi sheep disease virus (NSDV)). NSDV is found in large parts of East Africa and the Indian subcontinent (where it is known as Ganjam virus). We have investigated the ability of NSDV to antagonise the induction and actions of interferon. Both pathogenic and apathogenic isolates could actively inhibit the induction of type 1 interferon, and also blocked the signalling pathways of both type 1 and type 2 interferons. Using transient expression of viral proteins or sections of viral proteins, these activities all mapped to the ovarian tumour-like protease domain (OTU) found in the viral RNA polymerase. Virus infection, or expression of this OTU domain in transfected cells, led to a great reduction in the incorporation of ubiquitin or ISG15 protein into host cell proteins. Point mutations in the OTU that inhibited the protease activity also prevented it from antagonising interferon induction and action. Interestingly, a mutation at a peripheral site, which had little apparent effect on the ability of the OTU to inhibit ubiquitination and ISG15ylation, removed the ability of the OTU to block the induction of type 1 and the action of type 2 interferons, but had a lesser effect on the ability to block type 1 interferon action, suggesting that targets other than ubiquitin and ISG15 may be involved in the actions of the viral OTU.

Nairobi sheep disease virus (NSDV) is a member of the genus Nairovirus within the family Bunyaviridae and causes acute hemorrhagic gastroenteritis in sheep and goats, with very high morbidity and mortality rates in susceptible animals [1] . It was originally isolated in Nairobi, Kenya in 1910 by inoculation of sheep with the blood of sheep suffering from acute gastroenteritis. NSDV was originally thought to be endemic only in East Africa; recent sequence data showed that the same virus can also be found in many places in India and Sri Lanka where it is called Ganjam virus (GV) [2] . Daubney and Hudson showed that NSDV is primarily transmitted in East Africa by the hard tick Rhipicephalus appendiculatus, and that animals that were bred in areas where this tick was prevalent were immune, but animals that were moved into such areas died in large numbers [3, 4] . The virus is therefore only of limited effect on stable populations, but can be a severe limitation on trade or attempts to improve stocks through introduction of new animals. There is no current vaccine. In India, the virus is found in a number of tick species, primarily Hemaphysalis intermedia [5] . Sheep and goats are the only known vertebrate hosts of NSDV/GV [6, 7] , although one or two cases of human infection through needle-stick injury have been reported as leading to mild febrile illness [8, 9] . Nairoviruses are small, enveloped RNA viruses in which the genome consists of three segments of single stranded, negative sense RNA, designated Large (L), Medium (M) and Small (S) [10] . The S, M and L segments encode, respectively, the nucleocapsid protein (N), at least two envelope glyoproteins (Gn and Gc) and the viral RNA-dependent RNA-polymerase (L). The L segment is unusual among bunyaviruses, being extremely long (.12 kb), encoding a single protein of .450 kDa. The carboxyterminal half of this protein contains most of the polymerase motifs, while the amino-terminal part is largely of unknown function. The genus contains more than 30 different virus isolates, loosely grouped based on serum cross-reactivity and hemagglutination inhibition [11] , since sequence data on all but a few of these viruses has been limited or missing until recently. The most important serogroups are the NSDV serogroup, which also includes Dugbe virus (DUGV) and Kupe virus, and the Crimean Congo hemorrhagic fever virus (CCHFV) serogroup, which contains CCHFV and Hazara virus, both human pathogens. CCHFV causes a severe disease in human beings, with a reported mortality rate of 3-30% [12] . The disease is very similar to that caused in sheep by NSDV infection and is characterised by haemorrhage, myalgia, and fever. The first line of defence against virus infections is innate immunity. The key players are interferons (IFNs) and other cytokines that are rapidly produced in virus infected cells (Reviewed in [13] ). Three major classes of IFNs are known. Type I IFNs comprise the largest group, with multiple distinct IFNa genes, one to three IFNb genes and other genes (IFNv, -e, -d, -k). The first two are induced directly in response to viral infection whereas the others play less defined roles. Type II IFN has a single member, namely IFNc, which is secreted by activated T cells and natural killer cells rather than in direct response to virus infections. A third class of IFNs has been described recently that shares the same pathway to sense viral infection as type I IFNs and is also induced directly in response to viral infection [14, 15, 16] . After an infected cell senses a viral infection, IFNs are produced and released from the cell to induce an antiviral state in both itself and neighbouring cells. Both type I and type II IFNs bind to their cognate cell-surface receptors, thereby activating a signal-transduction pathway that triggers the transcription of several hundreds of genes [17] . These IFN-stimulated genes (ISGs) have either IFNstimulated response elements (ISRE) or GAS (Gamma-activated sequence) elements in their promoter region. Type I IFNs such as IFNa can lead to transcription of ISGs with ISRE or GAS elements, whereas IFNc can only induce ISGs with GAS elements (reviewed in [18] ). It seems that most viruses studied so far have developed a strategy to counteract the host innate immune system [13] . Two Nairoviruses (CCHFV, DUGV) have been shown to be inhibited by MxA, a protein induced by type I IFNs [19, 20] . Since the viruses are inhibited by IFN-induced proteins, viral virulence will depend at least in part on the ability of the virus to avoid or block the type I IFN response, and it is therefore very likely that Nairoviruses have developed tactics against this host defence mechanism. The current knowledge on how Nairoviruses manipulate the host innate immune system comes mostly from studies with CCHFV. Andersson et al. have shown that CCHFV showed a markedly delayed type I IFN response in cell culture, up to 48 hours after infection, possibly by interfering with the pathway that leads to activation of interferon regulatory factor 3 (IRF3) [21] . In addition they found that CCHFV is insensitive to IFNa treatment applied six hours post-transfection. One possible explanation for the delay in the IFN response is given by another study showing that CCHFV has developed a mechanism to remove the 59-terminal triphosphate group from its genome segments, thereby avoiding retinoic acid inducible gene I protein (RIG-I)-dependant IFN induction [22] . Another possible explanation may be the activity of the ovarian tumour-like (OTU) domain found in the amino-terminus of the viral L protein. Sequence analysis of the L proteins of Nairoviruses identified this domain, which represents a unique class of deubiquitinating enzymes (DUBs) [23] . Recent studies with the CCHFV OTUcontaining L protein showed that it decreases the coupling of ubiquitin (Ub) and the ubiquitin-like protein (Ubl) encoded by interferon stimulated gene 15 (ISG15) to cellular proteins [24] . The post-translational modification of proteins by Ub and Ubls regulates essential processes in the type I IFN response to viral pathogens [25, 26] . Cellular DUBs have been found which appear to act as part of negative feedback control systems for IFN induction pathways [27, 28] . Removal of Ub and/or ISG15 from their protein conjugates will disrupt a number of elements of the IFN induction pathway, and targeting the ISG15 and/or the ubiquitin system is a common strategy used by different viruses to inhibit innate immune responses [29, 30, 31, 32] . Up till now there is no information available on if and how NSDV/GV manipulates the host innate immune system. Because of the difficulties in handling CCHFV, there is only limited correlation of studies on virus and on viral proteins. We provide here the evidence that NSDV/GV is able to inhibit IFN induction as well as IFN action. We could identify the viral L protein as being responsible for these inhibitory effects. Furthermore NSDV is able to reduce total protein ubiquitination and ISG15ylation in infected cells. This deconjugating activity, as with CCHFV, is located in the N-terminal part of the NSDV L protein containing the OTU domain. Inactivation of the OTU enzymatic activity resulted in loss of its ability to antagonise IFN responses. Two isolates of NSDV/GV were available to us for these studies, one a highly tissue-culture passaged isolate of the virus from Uganda (and therefore notionally NSDV), the other an isolate from India (and therefore notionally GV) which had been passaged a limited number of times in mouse brain or BHK 21 cells. Although these two isolates have been shown phylogenetically to be the same virus [2] , we will refer to them as GV and NSDV in this paper to indicate the two isolates. The GV isolate proved to be still pathogenic in sheep, while the NSDV isolate was nearly completely attenuated, and the GV isolate is therefore probably nearer to wild type virus. As described in the Introduction, studies with CCHFV have shown it has a delayed IFN response in infected cells [21] . We wanted to know if NSDV/GV is able to interfere with the early induction of type I IFNs. To address this question we transfected Vero cells with a reporter gene construct containing the firefly luciferase under the control of the IFNb promoter. Luciferase expression is taken as a measure of protein production from this specific promoter after activation of transcription, which in turn is taken as a measure of IFNb induction. Although this system is commonly used to study the control of IFN induction, we recognise that failure to express luciferase could reflect viral effects at a number of different points in the induction, transcription and protein synthesis pathway. As a positive control for the induction of IFNb we used Newcastle disease virus (NDV), a paramyxovirus which has been reported as being unable to block IFN induction in mammalian cells [33] and therefore is frequently used as a model stimulator of cytoplasmic PRRs. Vero cells lack a functional IFN-b gene [34, 35] which made them a useful tool for our experiment because they allowed us to measure direct activation of the IFNb promoter in infected cells excluding any indirect effect of IFN synthesis in neighbouring uninfected cells. The transfected Vero cells were infected with the NSDV or GV isolate or with NDV and the amount of synthesised luciferase was determined at the indicated time points ( fig. 1a&b ). Infection with NDV induced a rapid activation of the IFNb promoter ( fig. 1a ); after 4.5 and 6.5 hours of NDV infection very high amounts of luciferase are already detected when compared to mock-infected cells, with levels of luciferase already decreasing at 9hpi. In contrast, at 4.5hpi there was no increase in promoter activity detectable in GV/ NSDV-infected cells when compared to mock-infected cells ( fig. 1b) . A very slight increase in luciferase was seen at 10hpi in NSDV-infected cells, and at 14 hpi there is a clear significant increase in the reporter gene activity induced by NSDV or GV infection with further increases up to 24 hours of infection. Recent evidence suggests that many negative-strand RNA viruses do not generate dsRNA during infection [36] and only induce IFN rapidly if they contain defective interfering particles (DIs) [37] . It is likely therefore that our NDV preparation is acting through a significant content of such DIs. Neither NSDV/GV isolate, however, induced interferon until later stages in the infection cycle. There was a significant difference between the two isolates, NSDV and GV, regarding the kinetics of activation of the IFNb promoter. Activation by the GV isolate was later than that induced by the NSDV isolate and also the transcriptional activation observed was lower when compared to the NSDV isolate ( fig. 1b ). This might be due to the fact that the NSDV isolate was previously passaged over 60 times in cell culture, in contrast to the GV isolate which has primarily been maintained in mouse brain culture and has been passed twice in BHK-21 culture in our hands; this difference may have resulted in the NSDV isolate having a reduced efficiency to evade IFN induction, or a more rapid growth in the cultured cells, leading to a more rapid production of one or more PAMPs recognised by the host cell. This second possible explanation is supported by the growth kinetics of GV and NSDV in Vero cells, which showed that NSDV replication rates were faster compared to GV, although both isolates grew to the same final titer (Lidia Lasecka, pers. communication). NSDV/GV is able to block transcription from the IFNb promoter at early stages of infection Next we wanted to examine if the absence of IFNb promoter activation in early stages of NSDV infection upon NSDV/GV infection is due to an active block or if NSDV/GV is rather avoiding the activation of the IFNb promoter by a similar mechanism as already described for CCHFV [21] . On that account we transfected Vero cells with the IFNb reporter gene plasmid. One day later cells were infected with the GV or NSDV isolate, and subsequently super-infected with NDV. Immunofluorescence experiments showed that approx 85% of cells were infected with NSDV/GV, and that prior infection with NSDV/ GV did not block NDV infection in these cells (data not shown). The reporter gene activity was determined as described in material and methods ( fig. 2 ). In cells infected with NDV alone the expected increase in the activity of the IFNb promoter was observed when compared to uninfected cells. Interestingly, when the NDV infection was preceded by infection with GV or NSDV there was a clear reduction in the promoter activation detectable compared to cells infected solely with NDV. The reduction in the effects of NDV in these assays was only 40% even though about 85% of the cells were infected with NSDV/GV. This may have been because there was insufficient NSDV/GV protein at the time of the NDV superinfection to provide a complete block of the IFN induction pathway inside each infected cell; the reduction in reporter gene activity was indeed less pronounced if NDV was applied after shorter periods of GV or NSDV infection such as four hours post infection (data not shown), showing that it takes some time for the active block of the IFN induction pathway to take effect. This discrepancy may also reflect the limited nature of the block provided by NSDV/GV, which is unable to completely prevent activation of the IFN-beta promoter later in its own infection (FIg 1b) , as well as the very strong stimulus provided by the NDV superinfection in these experiments. Nevertheless, these results indicate that NSDV and GV are able to actively suppress activation of the IFNb promoter, and are not simply avoiding detection by cellular PRRs. We wanted to know if NSDV is able to interfere with type I and type II IFN signalling. Vero cells were transfected with plasmids having the luciferase ORF under the control of the mouse Mx1 promoter (a promoter strongly activated by type I IFN) or one containing multiple copies of a GAS element (responds to type II IFN). Those cells were subsequently infected with either the GV or the NSDV isolate. After eighteen hours of infection cells were treated with IFNa or IFNc or left untreated. Finally the luciferase activity was determined as described in material and methods. Treatment of cells with IFNa induced high levels of luciferase activity in cells transfected with the Mx-1 reporter plasmid ( fig. 3a ) which is in accordance with the literature [38] . Infection with either GV or NSDV resulted in a significant reduction in IFNinduced Mx-1 promoter activity when compared to uninfected cells. The same effect could be observed when cells were treated with IFNc to induce the GAS element ( fig. 3b ). GV and NSDV were similarly effective in reducing the promoter activity in the presence of IFNc to less than forty percent of the activity in IFNcstimulated uninfected cells. These data strongly suggest that NSDV/GV is able to counteract the type I and II IFN induced transcription of genes that have an ISRE or GAS element in their promoters. The binding of IFNa/b to the type I IFN receptor results in the autophosphorylation and activation of JAK (Janus activated kinase) 1 and tyrosine kinase 2, which are both members of the JAK family and associated with the receptor (reviewed in [18] ). The activated JAKs phosphorylate specific tyrosines of STAT1 (signal transducer and activator of transcription 1) and STAT2. Upon phosphorylation, STAT1 and STAT2 form heterodimers and, in association with other factors, translocate to the nucleus to bind to ISREs to initiate transcription of ISGs. Binding of IFNc to its receptor similarly induces phosphorylation of STAT1 which then forms homodimers that translocate to the nucleus to activate the transcription of genes with GAS elements. Targeting the activation of STAT proteins or their transfer to the nucleus are efficient ways to block innate immunity that are used by several different viruses (reviewed in [13] ). To investigate whether NSDV/GV are inhibiting IFN action through an effect on STAT phosphorylation, we infected Vero cells with NSDV or GV for 16 hours and then stimulated the infected cells with IFNa or IFNc. Samples from those cells were subjected to immunoblot analysis to check the phosphorylation status of STAT1 and STAT2 ( fig. 4 ). In uninfected cells IFNa treatment induced phosphorylation of STAT1 and 2 ( fig. 4a ). An almost complete block of STAT1 phosphorylation was observed in NSDV infected cells, especially at later time points. The GV isolate also inhibited STAT1 phosphorylation, though was clearly less effective. This reduced effectiveness of the GV isolate corresponds to a reduced ability to block IFN action in the reporter gene assay ( fig. 3 ), and may be due to the slightly slower growth of this isolate in cell culture; at later time points of infection (18hpi) the GV isolate was The GV L protein inhibits transcription from the IFNb promoter So far the viral RNA-dependent RNA polymerase (RdRP) from CCHFV and DUGV are the only nairoviral proteins known to interfere with the host innate immune response [24] . We wanted to identify the NSDV protein(s) that is/are responsible for the inhibitory effects on IFN induction and action in infected cells. For that reason we made viral protein expression plasmids that were derived from the GV isolate as it is far less tissue culture adapted than the NSDV isolate. We cloned the open reading frame (ORF) of the S and M segments of GV into a mammalian expression vector under the control of a CMV promoter and with a V5-tag in frame at the C-terminus of each protein. For the M segment, which is translated into a polyprotein that produces at least two glycoproteins, this produces a V5 tag at the C terminaus of Gc, the most distal of the glycoproteins produced from this segment. We were unable to clone the complete ORF of GV (or NSDV) L into a plasmid due to recombination events that took place in Escherichia coli, even in strains such as STBL2, SURE2, ABLE K, XL-10 and MDS42 which are engineered to support unstable DNA. We mapped the toxic/unstable sequence to a region of approximately 1 kb found roughly in the middle of the L ORF. We were able to construct plasmids containing either half of this sequence but not the whole piece, and were therefore able to prepare expression constructs encoding the amino-and carboxy-terminal parts of GV L protein (aa L1-1757 and aa L1749-3391 respectively), thereby covering the whole protein with a short overlap between the constructs. In addition we cloned two shorter fragments of the amino terminal part of the viral polymerase that contained the OTU domain (aa L1-169) or the OTU domain and the following zinc-finger domain (aa L1-667). We transfected different amounts of these expression plasmids to approximately equalize the expression levels of the individual viral proteins in the reporter gene assays. In fig. 5a typical protein expression levels are shown that were used in our studies. Only the expression of the glycoprotein Gc which had the carboxy-terminal V5 tag could be analysed. The level of Gn could not be checked due to the lack of glycoprotein-specific antibodies for NSDV. However we assumed that the expression levels of the two glycoproteins are similar as they are expressed as a single polyprotein. We wanted to know if one of these viral proteins is able to block the transcription from the IFNb promoter in Vero cells infected with NDV. To address this question we co-transfected these viral protein expression constructs together with our reporter plasmids pIFNb-luc and pJATLacZ into Vero cells. Then the transfected cells were infected with NDV and luciferase activity in cell extracts was analysed ( fig. 5b ). Neither the nucleoprotein nor the glycoproteins could impede NDV-induced transcriptional upregulation of the IFNb promoter. All three amino-terminal expression constructs containing the OTU domain of the GV L protein (L1-169, L1-667 and L1-1757) were able to significantly decrease luciferase expression, whereas the C-terminal part of L (L1749-3391), which contains parts of the polymerase but no OTU domain, showed no effects on NDV-induced reporter gene expression. The protein L1-169 was more effective than L1-667 and L1-1757 in blocking NDV-induced IFNb-promoter activity. These data strongly suggest that the GV OTU domain is responsible for the inhibitory effects on type I interferon induction observed in infected cells (fig. 2 ). Our previous experiments have shown that NSDV is able to inhibit the action of type I and type II IFNs in infected cells ( fig. 3) . To identify the viral protein that is responsible for this effect we transfected Vero cells with reporter plasmids carrying a luciferase gene under the control of type I and II IFN-responsive promoters (pGL3-Mx-1-luc and GAS-luc) together with the viral protein expression plasmids. The cells were treated with IFNa or IFNc to induce transcription from the IFN-responsive promoters and the luciferase induced was measured ( fig. 6 ). The reporter gene activity in cells treated with IFNa or IFNc was significantly reduced in the presence of the GV L protein constructs containing the OTU domain (L1-169, L1-667 and L1-1757) compared to cells transfected with empty plasmid only, whereas in cells expressing the nucleoprotein, the glycoproteins or the carboxy-terminal part of the L protein, treatment with IFNa or IFNc induced comparable levels of reporter gene activity to that found in cells expressing no viral protein. As in the studies of IFN induction, the two shortest versions of the GV L protein were the most efficient in blocking the action of IFNa and IFNc. Data from studies with CCHFV L protein showed that the OTU domain is enough to inhibit total protein ubiquitination and ISG15ylation in 293T cells [24] . We wanted to know if NSDV/GV has any effects on total protein ubiquitination and/ or ISG15ylation in cells. We investigated this using Vero cells transfected with expression constructs for tagged forms of Ub or ISG15, with appropriate supporting plasmids as required for the ISG15 system [24, 39] . Cells were transfected with a plasmid expressing HA-tagged ubiquitin and subsequently infected with the GV or NSDV isolates. Twelve hours post-infection cell extracts were prepared and the expression of ubiquitin conjugated-proteins determined by immunoblotting ( fig. 7a) , showing that infection with either isolate caused a decrease in total protein ubiquitination compared to uninfected cells. In a similar way we studied the effects of NSDV/GV on ISG15ylation of proteins in infected host cells. ISG15 is a Ubl that is very rapidly induced in type I IFN-treated cells [40] and is thought to play important roles in anti-viral responses, either as a monomer or by conjugation to host cell proteins (reviewed in [41] ). ISG15 conjugation to host cell proteins exerts antiviral activity against influenza virus [42] and Sindbis virus [43] . For our studies on NSDV and its effects on host ISGylation during infection we made use of the fact that ISG15ylation can also be generated by transfecting expression constructs for ISG15 along with plasmids encoding the core components of the ISG15 conjugation system, the E1 activating protein (mUBE1L), E2 conjugating protein (UbcM8) and E3 ligase (mHerc6) into cells in the absence of IFN [39] . After transfecting these four plasmids into Vero cells, the cells were subsequently infected with the GV or NSDV isolates or left uninfected. Twelve hours post-infection cells were lysed and the level of ISG15-conjugates were examined by immunoblotting ( fig. 7b) . Infection of NSDV/ GV resulted in a drastic decrease in ISG15-conjugates found in Vero cells when compared to uninfected cells. Importantly, the infection with NSDV itself did not cause a reduction in the total amount of mono-ISG15 expressed from the transfected plasmid, as shown by tracks 5-7 of fig. 7b , where the helper plasmids were omitted, so one is simply comparing ISG15 expression in uninfected and infected cells. This excludes any indirect effect of infection on ISGylation through an effect on ISG15 expression or stability. Since the virus itself reduces Ub and ISG15 coupling to cell proteins, we wanted to confirm that the OTU domain is responsible for these decreased levels of conjugates in NSDV/ GV as it is for CCHFV. For these studies we used 293 cells, as those were the cells used in the studies on CCHFV proteins. We determined the level of ubiquitination in the presence of the GV L1-169 protein in 293FT cells and compared it to the level found in cells transfected with empty plasmid or with the CCHFV L1-354 protein (containing the OTU domain of CCHFV) (fig. 7c) . The CCHFV OTU-containing protein contains an HA tag and so appears in the same blots as the HA-Ub and HA-Ub conjugates. The CCHFV OTU completely abolished ubiquitination as previously shown [24] . The GV OTU was also effective in decreasing the levels of cellular Ub-conjugates, but was less effective compared to the CCHFV OTU domain. In the same way we investigated whether the OTU domain of NSDV/GV was responsible for the reduction in ISG15ylation. For this purpose we transfected 293FT cells with the core components of the ISG15 . 7d) . Expression of the GV OTU domain efficiently blocked conjugation of ISG15 to cellular proteins. The CCHFV OTU also blocked ISG15ylation in 293FT cells as already published [24] . Again the CCHFV OTU was more efficient than the GV OTU in reducing the amounts of ISG15conjugated proteins in the cell. Both virus isolates of NSDV are therefore able to decrease total cellular ubiquitination and ISG15ylation levels during infection. The GV OTU domain, when expressed alone, could reproduce these effects exerted by the virus during infection. To determine the role of the catalytic activity of the OTU domain in IFN antagonism, we changed the cysteine at position 40 and the histidine at position 151, two components of the catalytic triad, to alanine (C40A and H151A respectively). A third mutant was created where glutamine 16 was replaced by arginine (Q16R); Q16 has been described as important for the binding of the CCHFV OTU to ubiquitin but with little importance for the binding of ISG15 [44, 45] , so this mutation was designed to allow us to study solely the effects of deISG15ylation on the innate immune system. The mutations were introduced individually into the GV L1-169 construct, and transfected into 293FT cells to examine their effect on ubiquitination and ISGylation ( fig. 8a, b) . Mutation of C40 or H151 resulted in a complete loss of deubiquitinating ( fig. 8a) and deISG15ylating (fig. 8b) activity. The Q16R mutant performed like the wildtype regarding its ability to remove ISG15 from cellular substrates ( fig. 8b ) while its deubiquitinating activity was only slightly reduced when compared to the wildtype OTU ( fig. 8a) . In previous studies on CCHFV a similar mutant was described as being unable to hydrolyze a fluorogenic model DUB substrate (Ub-AMC) [44] . The difference might be explained by the fact that we are using the OTU domain from a different virus and also a different assay for DUB activity. We tested these mutants regarding their ability to interfere with IFN induction. For this purpose, cells were transfected with L1-169 or the catalytic mutants C40A, H151A or Q16R, along with the reporter plasmids pIFNb-luc and pJATLacZ ( fig. 8c) . The IFNb promoter was activated by infection of cells with Sendai virus (SV) for eight hours. Interestingly all three mutants lost their ability to block SV-induced IFNb promoter activity when compared to the wildtype. These results showed that the catalytic activity of the OTU is necessary to counteract IFN induction. The results with the Q16R mutant were interesting, as it retained its deubiquitinating and deISG15ylation activity but, despite that, was not able to block IFN induction. We also examined the ability of L1-169 C40A, H151A, and Q16R to block type I ( fig. 8d) and II (fig. 8e ) IFN action. We used the same experimental setup as described before when we tested the different viral proteins for their ability to block IFN action ( fig. 6 ). The catalytic site mutants of L1-169 no longer blocked type I IFN-induced gene expression (fig. 8c) . The Q16R mutant did block the induced transcription from the Mx-1 promoter to some extent, but not as efficiently as the wildtype. All mutants lost their ability to block type II IFN-induced gene expression (fig. 8d) . These data show that the enzymatic activity of the OTU domain in the nairovirus L protein plays a pivotal role in antagonizing IFN induction and action. In addition, the data from the Q16R mutant strongly suggests that specific, yet different, targets are involved in antagonising these three activities, since this protein is still able to remove ubiquitin and ISG15 from the dominant cellular substrates, but lost its ability to antagonize IFN induction and type II IFN action while retaining some ability to block type I IFN action. NSDV is regarded as one of the most pathogenic diseases in sheep and goats with mortality rates ranging from 40% in Merino sheep to 90% in Masai sheep [1] . Judging by its high pathogenesis, NSDV has most likely developed efficient mechanisms to circumvent or inhibit innate immunity. The first response of the immune system against virus infections is the production and secretion of type I IFNs. We could show that the induction of transcription from the IFNb promoter in infected cells by NSDV/ GV is delayed and reduced when compared to another negative strand RNA virus ( fig. 1) . We do not observe the extensive delay described previously in CCHFV infected cells [21] ; however, that observed in our studies would give the NSDV enough time to establish its infection and produce progeny virions (which takes approx 12 hours in Vero or BHK21 cells). A possible reason for the delayed induction could be a reduced production of PAMPs by the virus, such as the removal of the 59 triphosphate from progeny viral genome transcripts, as observed for CCHFV [21] . Interestingly our results showed that NSDV is able to actively suppress the induction of IFNb in infected cells, as both isolates were clearly able to reduce the NDV-induced transcription from the IFNb promoter by approximately 40% (fig. 2 ) when assayed at 8-12hpi, before either NSDV/GV isolate showed strong induction of IFN. Expression of the N-terminal part of the RNA-dependent RNA polymerase (L), which contains the OTU domain, was sufficient to reproduce the antagonistic effect on IFN induction in a reporter gene-based assay (fig. 5) , and we could show that mutations that affected the catalytic site of the OTU protease were no longer antagonists ( fig. 8 ). Ubiquitination and modification of cellular proteins with ubiquitin-like molecules such as ISG15 play important roles in regulating IFN induction through both the Toll-like receptor (TLR) and RIG-I-like receptor (RLR) pathways [46, 47] . Lys-63-linked polyubiquitination of RIG-I has been shown to be crucial for its ability to induce type I IFNs [48] . Arimoto et al. [49] showed that virus-induced IRF3 and NF-kB activation is dependent on the polyubiquitination of the protein NF-kB essential modulator (NEMO). Furthermore NF-kB activation is known to depend on ubiquitination of the inhibitor protein I-kB, targeting it for degradation [50] , while ISG15ylation enhances NF-kB activity by conjugating to and suppressing protein phosphatase 2Cb, which suppresses dephosphorylation of I-kB [51] . ISG15ylation positively regulates IRF-3 activation by preventing its interaction with PinI [52] . Wholesale removal of conjugated ubiquitin and ubiquitin-like modifiers would therefore greatly inhibit the IFN induction pathway. Indeed we observed a significant reduction in the levels of cellular ISG15-and ubiquitinconjugates during NSDV/GV infection ( fig. 7a, b) which could be attributed to the OTU domain of the L protein ( fig. 7c, d) . The enzymatic activity proved to be essential for the observed inhibition of IFN induction ( fig. 8e) . Several other viruses have exploited this mechanism of negatively regulating the IFN induction by encoding proteases of the OTU family [24, 30, 53, 54] . All these proteases exert deconjugating activities towards ubiquitin or the ubiquitin-like (ubl) molecule ISG15 or both. Interestingly, cells themselves make use of DUBs to regulate the IFN pathway but, in contrast to nairoviral OTUs, cellular DUBs tend to have highly specific targets [27, 55, 56, 57] . For example, deubiquitinating enzyme A (DUBA) selectively cleaved polyubiquitin chains on tumor necrosis factor receptorassociated factor 3 (TRAF3) and was identified in a small interfering RNA-based screen as a negative regulator of type I IFN production [27] . The crystal structure of the CCHFV OTU domain revealed a unique structure allowing it to bind ISG15 as well as ubiquitin, and this ability to interact with both conjugating proteins is one of the underlying reasons for the promiscuous activity of the nairovirus OTUs [44, 45, 58] . We also found that NSDV is able to inhibit the action of type I and II IFNs, and that this activity involves the inhibition of phosphorylation of both STAT1 and STAT2. The mechanism(s) of this inhibition remains to be determined. There is no reduction in the levels of STAT1/2, and an inhibition of their phosphorylation suggests either binding/sequestration of one or both STATs or inhibition of the IFN receptor-associated JAKs. We observed that IFNc-induced STAT1 phosphorylation was blocked similarly by both isolates at all times post infection, whereas IFNainduced STAT1/2 phosphorylation was dependent the degree of growth of the virus, the block developing more slowly in cells infected with the slower-growing GV isolate. This suggests that the mechanisms by which the virus exerts the blockade of type I and 2 IFNs are not the same. Further evidence for this came from the GV L1-169 Q16R mutant, which retained most of its DUB activity and its full deISG15ylating activity and still shows significant blockade of IFNa-induced gene expression but no longer blocks IFNc-induced gene expression ( fig. 8c, d) . It could be that this part of the N-terminus contains a specific-substrate binding site in addition to the catalytic core of the OTU domain, contributing to the effect of the OTU domain on specific cellular substrates; in this case the Q16R mutation exerts a steric inhibition on binding to the target involved in blocking type II IFN action, rather than an inhibition which is dependent on the OTU enzymatic activity. For the herpesvirus-associated ubiquitinspecific protease (HAUSP or USP7), an additional binding site with affinity for its target protein has been described in addition to its catalytic protease domain [59] . The virus had a stronger effect on IFNa-induced STAT1 phosphorylation than on STAT2 phosphorylation at all times, suggesting that STAT1 is the primary target. As with the block of IFN induction, the blockade of both type I and type II IFN actions mapped to the OTU domain of the L protein and required a functional OTU catalytic site. So far no definitive role for ubiquitination or ISG15ylation in type I or type II IFN signalling has been shown. JAK1 and STAT1 have been shown to be conjugated by ISG15 [60, 61] ; however, STAT1 phosphorylation in response to IFN is normal in cells from ISG15 knock-out mice [62] , suggesting that ISG15 plays no role in the immediate cell response to IFN. Differences were seen in the effectiveness of the different OTU-containing fragments of the GV L to block IFNa-or IFNcinduced gene transcription when expressed in our reporter genebased studies. GV L1-169 and L1-667 reduced the transcriptional activity induced by IFNa or IFNc to 30% and 40% respectively of the positive control, whereas the protein L1-1757 was less effective in reducing the transcriptional activity of the Mx-1 promoter or the GAS promoter (55% and 65% of the positive control respectively). These differences might reflect slight differences in the amounts of protein expressed in the transfected cells that cannot be detected by immunoblotting. Alternatively, the catalytic domain of the NSDV L protein might be autoinhibited by folding or oligomerisation in the longer construct L1-1757, in contrast to the shorter versions containing the OTU domain. A similar observation was made with the OTU domain containing nonstructural protein 2 (nsp2) from porcine reproductive and respiratory syndrome virus where a longer fragment was less effective in blocking NFkB promoter activity than a smaller version of this protein [30] . In addition it has been shown that CCHFV full length L displayed significantly less DUB activity than shorter versions of the protein [24] . Further crystal structures to extend that of the basic OTU domain [44, 45, 58] will clarify these points. One important factor that needs to be examined is the relationship of the reactivity of these OTU domains to species specificity of the viruses. Influenza B was found to inhibit the human but not the mouse ISG15ylation system [39] . The nairovirus OTUs appear to act by cleaving Ub and ISG15 from their respective conjugates, since mutations in the active site abolish activity [24] , and the OTUs are therefore active against both human and murine systems. However, there was a clear difference between the activity of the CCHFV and NSDV/GV OTUs in the murine ISG15ylation system used in this study. We know little of the ruminant equivalents of the Ubls, and it is possible that the OTUs of different viruses may be adapted to species differences in these modifying proteins which could lead in turn to differences in the species specificity of pathogenesis. It will be important to examine the actual level of modification of host cell proteins when these viruses are grown in cells from different hosts to establish whether there is any degree of correlation between OTU cleavage activity and host cell species. Within this study we could demonstrate that NSDV/GV is able to block the innate immune system at three different stages, type I IFN induction, type I IFN action and type II IFN action. NSDV/ GV seems to be another example of a pathogen that exploits the host cell ubiquitin pathways for its own good by encoding an enzyme with deubiquitinating and deISG15ylating activity. However, to fully clarify the role of the OTU activity, an OTU knock-out virus is needed to evaluate the importance of the OTU domain in vivo, and we are working towards such a system. An advantage of NSDV/GV will be the ability to carry out such experiments in the natural host. The Vero cells (African green monkey kidney cells) used in these studies were a modified line that expresses CD150 (aka Signaling Lymphocyte Activation Molecule (SLAM), as these were the Vero Except where indicated, all DNA manipulation was done following standard methods. Plasmids were cloned and grown in Escherichia coli DH5a or SURE2 (Stratagene). Routinely plasmid DNA was purified on CsCl gradients. The plasmids pJAT-lacZ, pGAS-luc, and pIFNb-luc were the kind gifts of Prof Steve Goodbourn, St. George's Hospital Medical School, London, United Kingdom. The pGL3-Mx1P-luc was kindly provided by Prof Georg Kochs, Department of Virology, University of Freiburg, Germany. The following plasmids were used for the ISG15ylation and ubiquitination experiments: pCAGGS.MCS-6HismISG15, and plasmids expressing mHerc6, Ubcm8 and mUbE1L were provided by Prof Deborah J. Lenschow, Washington University School of Medicine, St. Louis, Missouri. Plasmid pHA-CCHFV-L1-354 is in pCAGGS-MCSII and was the gift of Dr Natalia Frias-Staheli, Mount Sinai School of Medicine, New York. Construction of viral protein expression plasmids. Total RNA from GV-infected BHK21 cells was extracted by using RNeasy Mini Kit (Quiagen) which served as a template for cDNA synthesis using random priming oligos and SuperscriptII Reverse Transcriptase (Invitrogen). The viral genome was amplified by PCR using NSDV/GV genome-specific oligos and subsequently blunt-end cloned into pT7Blue (Novagen). All PCRs were performed using proofreading polymerase (KOD; Novagen). The genome of both strains was completely sequenced. Plasmids pcDNA-GV-N and pcDNA-GV-M were made by cloning the complete ORFs of GV S and GV M segments into pcDNA6/V5-His (Invitrogen), expressing C-terminal V5-tagged N and the glyoproteins (only Gc has a V5 tag at its C-terminus). GV L1-169 and L1-1757 were cloned in pcDNA6 with a C-terminal V5 tag. GV L1-667 and L1749-3391 were cloned into pTriEX (Novagen) such that the expressed proteins have a 6xHis tag at the C-terminus. To generate catalytically inactive variants of L1-169 in pcDNA6/V5-His, single amino acid mutations were introduced by overlap PCR mutagenesis. All mutations were confirmed by sequencing the entire open reading frame. Mouse monoclonal antibody against phosphotyrosine 701-STAT1 were purchased from BD Biosciences. Polyclonal antibodies against STAT1, STAT2, and phosphotyrosine 689-STAT2 were obtained from Upstate. Mouse monoclonal antibody against proliferating cell nuclear antigen (PCNA) was obtained from Santa Cruz Biotechnology. Mouse monoclonal anti-His antibody was purchased from Sigma Aldrich and HRP-tagged anti-HA antibody from Roche. The rabbit anti-N antiserum recognizing the amino terminus of the NSDV and GV N protein were previously made in our laboratory. Mouse monoclonal antibody against the V5 epitope tag was purchased from AbD Serotech. Mouse monoclonal antibody U85 recognising Newcastle disease virus was the kind gift of Ruth Manvell, AHVLA, Weybridge, UK. All transfections were carried out with TransIT LT1 (Mirus) according to the manufacturer's instructions. A ratio of 2 or 3 ml LT1 per mg DNA was used. Cells were plated at 10 5 per well in 12well plates 1 day before use. Usually, 24 hours post transfection the medium containing the transfection mix was removed and replaced with fresh medium. The cells were then lysed in 200 ml lysis buffer (120 mM NaCl, 50 mM TrisCl pH 7.5, 0.05% Nonidet P-40). The samples were centrifuged for 1 min at full speed in a table top centrifuge. 50 ml of the cleared cell extract was taken and the luciferase activity was measured after adding 50 ml of luciferase assay reagent (Promega) to each sample. The following settings were used to measure the relative light units: integration time: 10 sec, sensitivity: 200 or 230 and Filter set 1. To determine the b-galactosidase activity 150 ml of assay buffer (48 mM Na 2 HPO 4 , 32 mM NaH 2 PO 4 , 8 mM KCl, 0.8 mM MgSO 4 , 3.2 mg/ml o-Nitrophenyl b-D-Galactopyranoside) were added to the samples after the luminescence was measured. The samples were incubated at 37uC for 30 min to 60 min before absorbance at 420 nm was measured. The luminescence and absorbance measurements were done in a Synergy 2, Multi Detection Microplate Reader (BioTek Instruments) using Gen5 software. The values were normalised and statistically analysed as previously described [63] . Vero cells were plated at an initial seeding density of 1610 5 / well in 12-well plates. One day later cells were infected with GV or NSDV at a multiplicity of infection (MOI) of 1. The virus inoculum was removed one hour after infection, cells were washed once with PBS and fresh medium was added. The infected cells were further incubated for 14 h before being treated with 1000 IU/ml recombinant human aA-Interferon (IFN) or 1000 IU/ml recombinant human IFN-c. IFN-aA was purchased from Calbiochem and IFN-c was obtained from Millipore. Cells were treated for 30 min with or without IFN before being harvested and lysed with 100 ml of 1x SDS sample buffer (New England Biolabs). SDS-PAGE and Western blots were carried out as previously described [64] .

A Human Monoclonal Antibody with Neutralizing Activity against Highly Divergent Influenza Subtypes

The interest in broad-range anti-influenza A monoclonal antibodies (mAbs) has recently been strengthened by the identification of anti-hemagglutinin (HA) mAbs endowed with heterosubtypic neutralizing activity to be used in the design of “universal” prophylactic or therapeutic tools. However, the majority of the single mAbs described to date do not bind and neutralize viral isolates belonging to highly divergent subtypes clustering into the two different HA-based influenza phylogenetic groups: the group 1 including, among others, subtypes H1, H2, H5 and H9 and the group 2 including, among others, H3 subtype. Here, we describe a human mAb, named PN-SIA28, capable of binding and neutralizing all tested isolates belonging to phylogenetic group 1, including H1N1, H2N2, H5N1 and H9N2 subtypes and several isolates belonging to group 2, including H3N2 isolates from the first period of the 1968 pandemic. Therefore, PN-SIA28 is capable of neutralizing isolates belonging to subtypes responsible of all the reported pandemics, as well as other subtypes with pandemic potential. The region recognized by PN-SIA28 has been identified on the stem region of HA and includes residues highly conserved among the different influenza subtypes. A deep characterization of PN-SIA28 features may represent a useful help in the improvement of available anti-influenza therapeutic strategies and can provide new tools for the development of universal vaccinal strategies.

Influenza, one of the diseases that has shaped human history [1, 2] , still has an evident clinical and socio-economical impact [3, 4] . The 2009 pandemic has raised several major concerns related to the few prophylactic and therapeutic measures available. Antiviral compounds have drawbacks caused by the rapid emergence of drug-resistant isolates [5, 6] , require prompt administration to be effective [7] , and have several associated side-effects especially in high-risk categories, including children and pregnant women [8, 9] . Additionally, the vaccinal strategy is exposed to the annual risk of being ineffective due to possible mismatches between the predicted strains included in the vaccine and those actually in circulation; moreover, it would not engender a prompt response in pandemic settings [10] . In this scenario, new broadrange ''universal'' anti-influenza strategies are required [11, 12] . In particular, it would be important to identify and eventually elicit what has recently been described as an unusually extreme broadrange immunity directed against broadly conserved viral regions, differing from the more common and restricted immunity directed against highly variable regions [12] . A number of approaches have already been proposed in literature [10, 12, 13, 14, 15, 16] , but a pivotal role, both in the prophylactic and therapeutic field, may be played by the availability of broad-range neutralizing human monoclonal antibodies (mAbs) allowing the identification of human B epitopes widely shared among different influenza subtypes [11, 12] . Indeed, it is accepted that antibodies are key players in natural protection against influenza viruses, and that hemagglutinin (HA) is the main target for the virus-neutralizing antibody response [17] . However, although a single influenza infection provides lifelong immunity against the infecting virus and a limited number of antigenically correlated strains, the host can remain susceptible to infection with an antigenically drifted variant due to HA variability [18] . HA is the major glycoprotein of the influenza virus; it binds sialic acid on the surface of the cells through its globular head (HA1 domain) and makes possible the fusion of the viral envelope with the endosomal membranes through its stalk region (mainly formed by the HA2 domain) [17] . The sixteen known subtypes of HA, sharing between 40% and 60% amino acid sequence identity, have been clustered in two distinct phylogenetic groups: group 1 (H1, H2, H5, H6, H8, H9, H11, H12, H13, and H16) and group 2 (H3, H4, H7, H10, H14, and H15) [12, 14] . The subtypes, H1, H2, H5 and H9 in group 1, and H3 and H7 in group 2 have been isolated in humans and in particular H1, H2 and H3 subtypes have been responsible of the reported influenza pandemic outbreaks. In this study, we describe a human mAb, named PN-SIA28, that is capable of neutralizing all tested group 1 isolates, as well as isolates belonging to H3N2, the only group 2 subtype capable, so far, of causing a pandemic. The binding and neutralizing features of PN-SIA28 were initially studied using the Fab fragment molecule produced in E. coli, demonstrating that Fab PN-SIA28 recognizes an epitope on the stem region of HA and is able to strongly neutralize all tested H1N1 strains [19, 20] . It is well documented in the literature that bivalency of a whole IgG molecule may be an essential features for the biological activity of a mAb [21, 22, 23, 24] . For this reason, in this work, to evaluate PN-SIA28 features as whole IgG molecule, IgG PN-SIA28 was generated and tested in different neutralization assays against human, swine and avian influenza A viruses belonging to both HA based phylogenetic groups and encompassing all subtypes responsible of described pandemic events. The results obtained showed that IgG PN-SIA28 strongly neutralizes viruses belonging to the group 1 as well as those of the group 2. More in details, IgG PN-SIA28 neutralized all the H1N1 tested viruses with an half maximal inhibitory concentration (IC 50 ) ranging between 0.4-3.7 mg/ml, the H5N1 viruses with IC 50 ranging between 0.9-2.8 mg/ml, the H2N2 subtype isolate with an IC 50 of 0.8 mg/ml and the H9N2 subtype strain with an IC 50 of 0.9 mg/ml (Table 1, Figure S1 ). More interestingly, IgG PN-SIA28, also potently neutralizes H3N2 viruses circulating in the years from 1968 to 1975 with IC 50 ranging between 0.8-2.6 mg/ml (Table 1, Figure S2 ) despite the great phylogenetic distance of these viruses compared to those belonging to group 1 ( Figure 1 ). No detectable neutralizing activity was observed against more recent H3N2 tested viruses (Table 1, Figure 1 and S3) as well as for the H7N2 tested isolate. Screening of a random 12-mer peptide phage displayed library. In order to select peptides able to bind antibody PN-SIA28 that could possibly indicate the region that the antibody recognizes on the HA, a phagemidic random 12-mer peptide library was screened. After three rounds of panning, PN-SIA28 binding phages were eluted and used to infect E. coli XL1-Blue for an ELISA screening on PN-SIA28. Twenty-five positive clones were sequenced. An in silico analysis was performed, using Mimox and Pepitope servers, comparing the selected peptides to the available crystal structure of H1N1 A/South Carolina/1918 (A/ SC/1918) (ID code 1RD8) and A/Puerto Rico/8/34 (A/PR/8/ 34) hemagglutinins (ID code 1RU7). This study allowed the identification of several possible residues located on the stem region of HA potentially involved in the binding of PN-SIA28: Asn336, Ile337, Pro338, Trp357 and Thr358 (sequence numbering refers to A/PR/8/34, GenBank accession number ABO21709) ( Figure 2 ). In vitro selection of escape mutants. To further define the region bound by PN-SIA28, an assay aiming at the evaluation of PN-SIA28 capability to induce escape mutant was performed. A/ PR/8/34 H1N1 virus was cultured under constant selective pressure of PN-SIA28, and after several rounds of cell infections, two escape mutants were generated. Sequencing analysis of the generated neutralization escape variants revealed two different mutants carrying each a single amino acidic mutation in position 361 (Ile361Thr) and in position 362 (Asp362Gly) in the HA2 subunit compared to wild type. Alanine scanning mutagenesis study. On the basis of these results, HA mutants carrying an alanine substitution in position 361 (Ile361Ala) or in position 362 (Asp362Ala) were generated and the PN-SIA28 binding to the mutants was evaluated by FACS analysis evidencing that PN-SIA28 was not longer able to bind to the mutated HAs. In order to identify other amino acidic residues involved in the binding of PN-SIA28 to HA, a large panel of HAalanine mutants was generated (Table S1 ). FACS analysis showed that binding of PN-SIA28 to HA was decreased by His25Ala, His45Ala mutants on HA1 and Thr358Ala, Met360Ala, Ile361Ala, Asp362Ala, Gly363Ala, Trp364Ala, Thr384Ala and Val395Ala mutants on HA2. (Figure 2 ). All other mutations listed in the Table S1 did not have any effect on PN-SIA28 binding. An extra HA-mutant in position 361 (Ile361Val) was generated. Indeed, in position 361 either an Isoleucine or a Valine residue can be present in different isolates belonging to either to group 1 either to group 2 ( Figure 2 ). No binding reduction of PN-SIA28 to this mutant was observed. Based on the results obtained with the HA mutants, an in silico analysis on the HA crystal structure of A/SC/1918 and A/PR/8/ 34 was carried out for the amino acid residues that the alanine scanning study showed to influence the binding of PN-SIA28 to HA. The analysis confirmed that the residues identified lie on the stem region of HA, that they belong to the HA1 and HA2 subunits and that they are exposed on the surface of the HA molecule ( Figure 3) . A similar approach was followed also for the H3 subtype. Based on the results obtained for the H1N1 study, four H3 alanine mutants were initially generated on the A/Aichi/2/68 (H3N2) HA: His34Ala, Asn54Ala on HA1 and Ile363Ala and Asp364Ala on HA2 (corresponding to His25, His45, Ile361 and Asp362 in H1N1 sequencing numbering) ( Figure 2 , Figure 4 and Table S1 ). None of these alanine mutated HAs was bound by PN-SIA28. Furthermore, the Asn54His mutant (corresponding to position 45 in H1N1 numbering) on A/Aichi/2/1968 HA was also generated due to the presence of a Histidine in the corresponding position in H1N1 viruses. An increased binding of PN-SIA28 to this mutant was observed compared to wild type H3 HA. In order to deeper investigate on the absence of neutralizing activity of PN-SIA28 against most recent H3N2 strains, three more alanine-mutants (Asp18, Lys57 and Ile70) (see Table S1 ) were generated considering the differences in the stem region between H3N2 A/Victoria/3/1975 (the most recent neutralized isolate) and H3N2 A/Philippines/2/1982 (the earliest non neutralized strain). A 30% binding decrease was observed for Lys57Ala HA mutant, while the Asp18Ala and Ile70Ala did not have any effect on PN-SIA28 binding to H3N2 HA ( Figure 5 and Table S1 ). In the present study we describe the binding features and the neutralizing activity of a human monoclonal antibody named PN-SIA28 previously described as Fab fragment [19, 20] . IgG PN-SIA28 was tested against a large panel of influenza A viruses belonging to group 1 (H1N1, H2N2, H5N1, H9N2) and group 2 (H3N2, H7N2) subtypes showing a broader neutralizing activity compared to its monovalent molecule, possibly due to the bivalency feature of the IgG. Indeed, IgG PN-SIA28 showed a Figure 1 . PN-SIA28 Neutralizing activity. Influenza hemagglutinin unrooted phylogenetic tree of the different viral strains tested in neutralization assays with PN-SIA28. Viral isolates belonging to group 1 and group 2 are presented in the box B and box A, respectively. A green '+' indicates positive neutralizing activity, a red '2' indicates negative neutralizing activity. As reported in the text, PN-SIA28 is able to neutralize all of the group 1 strains and is also able to neutralize all of the H3N2 isolates spanning 1968 and 1975. *Recombinant HA from (H1N1) A/South Carolina/1/ 1918 pandemic strain was previously shown to be bound by PN-SIA28 [19, 20] . Analogously, recombinant HA from H5N1 A/Cygnus Olor/Italy/742/ 2005 was recognized by PN-SIA28 (data not shown). # H1N1 A/New Caledonia/20/1999 was previously shown to be neutralized by PN-SIA28 as Fab fragment [19, 20] . doi:10.1371/journal.pone.0028001.g001 robust neutralizing activity (IC 50 = 0.4-3.7 mg/ml) against all tested strains belonging to group 1, thus demonstrating that its epitope is broadly shared among group 1 viruses, including the highly divergent H9 subtype. More importantly, PN-SIA28 showed a potent neutralizing activity also against group 2 viruses demonstrating that its epitope is also present in the highly Figure 2 . Sequence conservation in hemagglutinin groups and subtypes. Boxes indicate mutated residues which decrease PN-SIA28 binding to mutated HAs. Circles on the top indicate PN-SIA28 percent binding to each HA alanine mutants compared to binding to wild-type HA: red 25% binding, yellow 50-75% binding. Sequence numbering is based on H1N1 A/PR/8/34 coding region (GenBank accession number ABO21709). Neutralizing activity of PN-SIA28 against each strain is highlighted by a green '+' or a red '2' on the left, indicating neutralizing activity and no neutralizing activity, respectively. *Recombinant HA from H1N1 A/South Carolina/1/1918 pandemic strain was previously shown to be bound by PN-SIA28 [19, 20] ; analogously, recombinant HA from H5N1 A/Cygnus Olor/Italy/742/2005 was recognized by PN-SIA28 (data not shown). # H1N1 A/New Caledonia/20/1999 was previously shown to be neutralized by PN-SIA28 as Fab fragment [19, 20] . doi:10.1371/journal.pone.0028001.g002 divergent H3N2 subtype. Indeed, IgG PN-SIA28 was able to potently neutralize H3N2 strains circulating from 1968 to 1975 (IC 50 = 0.8-2.6 mg/ml), despite the important phylogenetic distance between these viruses and those belonging to group 1 ( Figure 1 ). No neutralizing activity was observed against the dramatically divergent H7N2 isolate tested in this study, nor against the more recent H3N2 isolates. The broad neutralizing activity of PN-SIA28 against highly divergent influenza viruses, suggests that the epitope recognized is extremely conserved. The lack of hemagglutination inhibitory activity and the competition with a mouse mAb (C179) directed against the stem region of HA, had already suggested that PN-SIA28 epitope is not localized on the globular head and that it is shared between HA1 and HA2 domains [19, 20] . Several approaches have been used in this study to better define the region recognized by PN-SIA28. The screening of a phagemidic random peptide library suggested the binding of PN-SIA28 to the stem region in the proximity of the viral membrane. Interestingly, the sequence analysis of A/PR/8/34 (H1N1) escape mutants, generated under PN-SIA28 selective pressure, identified two residues (Ile361 and Asp362) localized in close proximity to the residues identified with the peptide library approach. It is worth noting that none of these two specific mutations is present in any of the more than 6,000 sequences available in public influenza databases (http://www.ncbi.nlm.nih.gov/genomes/FLU/FLU.html), suggesting that the immune pressure on these residues is poor and that the region is functionally highly conserved in all subtypes (rate of non-conservative substitutions: 0.17% for residue 361, and 8.7% for residue 362). Based on these data, A/PR/8/34 (H1N1) HA mutants carrying an alanine substitution in position 361 (Ile361Ala) or in position 362 (Asp362Ala) were generated and the FACS analysis showed that PN-SIA28 lost most of its capability to bind the mutated HA, in concordance with the data obtained from the escape mutants. The residue in position 361 was also mutated in Valine (Ile361Val), an aminoacid largely present in most of the neutralized isolates (Figure 2 ), indeed confirming that the presence of this residue did not affect binding. With the help of available HA crystal structures, a larger panel of HA-A/PR/8/34 (H1N1) alanine mutants was analogously generated and several residues (His25Ala, His45Ala on HA1; Thr358Ala, Met360Ala, Ile361Ala, Asp362Ala, Gly363Ala, Trp364Ala, Thr384Ala and Val395Ala on HA2), close to those identified with the escape mutants approach, influenced the interaction of PN-SIA28 with HA ( Figure 2 and 3) . The more restricted panel of alanine mutants (His25Ala, Asn45Ala, Ile361Ala and Asp362Ala) generated for A/Aichi/2/68 (H3N2) HA, confirmed the key role of some of the residues already identified as implicated in the interaction of PN-SIA28 with HA (H1) (Figure 4 ). Three more mutants were generated considering the differences among solvent exposed residues close to PN-SIA28 epitope-core on HA stem region between A/Victoria/3/1975 (the most recent neutralized isolate) and A/Philippines/2/1982 (the earliest non neutralized strain) isolates. The only mutant partially affecting PN-SIA28 binding was the one on Lys57, a residue present among all H3N2 neutralized strains and substituted by several different residues among H3N2 not neutralized strains. This could partially explaining the lack of neutralizing activity against most recent strains ( Figure 5 and Table S1 ). Overall, these results indicate that the region bound by PN-SIA28 encompasses residues on the HA stem, and that it is partially overlapping from the binding region of other human neutralizing antibodies with heterosubtipic neutralizing activity limited to influenza A group 1 viruses [25, 26, 27, 28, 29] or group 2 viruses [30, 31] . In addition, the region bound is also partially shared with FI6v3, an optimized human mAb able to crossneutralize influenza A group 1 and group 2 viruses [32] . Importantly, both the heavy chains of PN-SIA28 and FI6v3 derive from rearrangement of the same VH gene germline (VH3-30). As already hypothesized for the VH1-69 subfamily, widely shared among human mAbs with heterosubtypic neutralizing activity limited to group 1 viruses, the VH3-30 gene structure may be important in conferring the broader heterosubtypic activity observed for PN-SIA28 and FI6v3. The future definition of HA/ PN-SIA28 crystal structure and its comparison with HA/FI6v3 complex may help confirm the possible role of conserved VH3-30 residues in conferring such unusual neutralization properties. Although the resolution of the crystal structure will be necessary for the fine definition of the PN-SIA28 epitope, several important considerations may already be made. Firstly, it was previously speculated that the presence of an additional glycosilation site (Asn54 corresponding to His45 present on group 1 viruses) on the HA1 portion of the stem region in H3N2 viruses is the reason for the lack of anti-H3N2 activity of other mAbs with heterosubtypic activity limited to group 1 subtypes [25, 26, 30] . The present study demonstrates that this extra glycosilation site on H3N2 does not preclude PN-SIA28 binding and neutralizing activity against viruses belonging to this subtype ( Figure 2 ). The testing of HA alanine mutants for position His45 on H1N1 and Asn54 on H3N2, and the increased binding of PN-SIA28 to Asn54His mutant on H3N2, confirm that this key residue is particularly important for the interaction of PN-SIA28 with HA. However, other residues, including Lys57 and others not yet identified, certainly play an important role in the PN-SIA28 binding to HA, as shown by the lack of activity observed against the recent H3N2 isolates tested, as well as against the H7N2 isolate, despite a substantial identity in the studied region ( Figure 2 ). In conclusion, the molecule described in this paper is a singular human mAb featuring a broad heterosubtypic neutralizing activity encompassing group 1 and group 2 influenza A subtypes, and therefore recognizing a broad-range neutralizing human B epitope in the stem region of HA. Indeed, PN-SIA28 neutralizes influenza viruses belonging to all subtypes that have caused pandemics in humans, as well as other subtypes with pandemic potential. The data presented may therefore be crucial not only for the improvement of classical passive anti-influenza prophylactic and therapeutic strategies [33, 34] , but also for the correct understanding of mechanisms leading to in vivo protective anti-influenza immunity and therefore for the design of more effective vaccinal strategies. Indeed, several pharmacokinetic studies have shown that the extremely low IC 50 featured by PN-SIA28 against most of the isolates tested can easily be reached in humans after systemic administration [35, 36] . Given the frequent correlation between a low IC 50 value and in vivo efficacy observed for other anti-influenza mAbs [25, 26, 27, 28, 29, 30, 31, 37, 38, 39] , PN-SIA28 could be a really promising compound to be used in passive immunization strategies. More importantly, PN-SIA28 represents the molecular evidence that an extremely wide unusual neutralizing immunity although, uncommon, may be elicited during the course of a natural infection and, potentially, even after a new-generation vaccinal approach focused on its epitope [40] . As a consequence, PN-SIA28 could be crucial in the identification of promising HAbased vaccinal strategies aimed at the elicitation of a broadly heterosubtypic protective immunity similar to the one represented by this human mAb. As described for a few other mAbs [25, 26, 27, 28, 29, 30, 31, 32, 41] , PN-SIA28 is mainly directed against the highly conserved HA2 domain whose use in novel anti-influenza vaccinal approaches is being widely investigated. However, PN-SIA28 demonstrates that not all antibodies directed against the stem region have the same characteristics, and that their neutralizing activity is modulated by the specifically recognized epitope. Indeed, the data presented in this paper show that in the design of such innovative approaches (i.e. the so-called ''headless approach'' [16] ) it is important to identify also residues belonging to HA1 domain to be preserved to elicit the widest immunity. Several immunogens are usually obtained on the basis of in silico analysis, but their real protective potential can only be ascertained after time-consuming and expensive in vivo studies. Under this perspective, PN-SIA28 could therefore be an extremely important reagent to identify immunogens more likely to stimulate antibodies with a similar broad heterosubtypic neutralizing activity that will constitute the best candidates for further in vivo studies. The isolation of PN-SIA28 has been previously described [19, 20] . In brief, PN-SIA28 was isolated from a 55 year old patient with a negative clinical history of infection from influenza viruses during the past ten years, and with a detectable serum neutralizing titre (half maximal inhibitory dilution 2 ID 50 . = 1:20) against two reference strains belonging to two distinct HA-based phylogenetic groups: the H1N1 strain A/Puerto Rico/8/1934 for group 1 and the H3N2 strain A/Port Chalmers/1/1973 for group 2. It was then expressed in E. coli as Fab fragment to define its features. To obtain the whole IgG described in this study, the BD BaculoGold System (BD Biosciences Pharmingen, San Diego, CA, USA) was used. Briefly, nucleotide sequences codifying heavy and light chains of PN-SIA28 Fab fragment were sub-cloned into the baculovirus expression vector pAc-k-Fc (PROGEN Biotechnik GmbH, Heidelberg, Germany). Sf solution containing the antibody was dialyzed against PBS and then concentrated using Amicon Ultra-15 Centrifugal Filter Devices (Millipore, Billerica, MA, USA). Antibody concentration was determined by SDS-PAGE gel and by spectrophotometric measurement at 280 nm. An anti-influenza A antibody directed against the HA (H1N1 subtype), named RB62, and an anti-HCV E2 glycoprotein antibody, named e137, produced and purified with an identical procedure were used as controls in all experiments. The following reference strains were tested in the BLS3 laboratory of the Vita-Salute San Raffaele University: (H1N1 . The A/swine/Parma/1/97 isolate was analogously grown on NSK (Newborn Swine Kidney) cells, kindly provided by the Zooprophylactic Institute of Brescia, Italy. At 80% confluence, cells in MEM supplemented with 2 mg/ml serum-free TPCKtrypsin (Roche Applied Science), were infected with each strain at a MOI of 0.001. After 1 hour of infection, cells were washed with PBS (Phosphate buffered saline); MEM supplemented with 2 mg/ ml trypsin was then added and cells were incubated at 37uC in 5% CO 2 atmosphere. Cells were observed daily to monitor the cytopathic effect and, usually after 72-96 hours, the supernatant was collected, centrifuged at 1000 rcf for 10 minutes to eliminate cells debris and filtered with 0.22 mm filters (Millipore, Billerica, MA, USA). The supernatant was then aliquoted and stored at 280uC as cell-free virus stock. Fluorescence inhibition assay. Each viral isolate was titrated by the limiting dilution method and the viral titer calculated by the Reed-Muench formula. Neutralizing assays were carried out in 96 wells plate using MDCK cells (46104cells/ well). Serial dilutions, 30 mg/ml-0.03 mg/ml, of IgG PN-SIA28 were preincubated for 1 hour at 37uC with 100 TCID50 of each H1N1 or H3N2 virus. Following incubation, 100 ml of the mix antibody-virus were added to the cells and incubated for another hour at 37uC in 5% CO2. At the end of this incubation, cells were washed with PBS and 100 ml of MEM TPCK-Trypsin (2 mg/ml) were added in each well. Cells were incubated for 7 hours at 37uC in 5% CO2 and then washed with PBS, fixed and permeabilized with ice-cold ethanol. Cells were incubated with anti-influenza A mouse antibody (Argene, Shirley, NY, USA) for 30 minutes at 37uC in a humid chamber. The cells were then washed and incubated for 30 minutes at 37uC in a dark humid chamber with a FITC-conjugated secondary antibody (Argene, Shirley, NY, USA). Nuclei staining was obtained with Hoechst 33342 (Sigma Aldrich). An infection control without antibody was included, as well as a negative control with an anti-HCV/E2 antibody (e137). Each neutralization assay was performed in triplicate and repeated in two different sessions. The neutralization activity for each antibody concentration was expressed as the percentage reduction of fluorescent nuclei compared with the nuclei count in the infection control. Nuclei counting was performed by using the GE Healthcare's IN Cell Analyzer 1000, an automated epifluorescence based microscope system. The neutralization curves were then fit by non-linear regression with the GraphPad Prism software, allowing IC 50 calculation. Colorimetric assay. Each viral isolate was titrated to establish working dilution that produces 15-30 foci forming units per well in 96 tissue culture plates. Neutralizing assays were carried out in 96 wells plate using MDCK/SIAT-1 cells. Serial dilutions, 30 mg/ml-0.37 mg/ml, of IgG PN-SIA28 were preincubated for 1 hour at 37uC with the subset of viruses. Following this incubation, 100 ml of the antibody-virus mix was added to the cells and incubated for another hour at 37uC in 5% CO 2 . At the end of this incubation, the cells were washed twice in PBS and 100 ml of virus growth media containing 2 mg/ml of TPCK treated trypsin was added. Cells were incubated for 12-16 hours at 37uC in 5% CO 2 and then washed with PBS, fixed and permeabilized with ice cold methanol/acetic acid (95:5) for 30 min at 220uC. Cells were incubated with anti-NP antibodies (Millipore, Billerica, MA, USA) for 30 minutes at 37u. The cells were then washed and incubated for 30 additional minutes at 37uC with a mouse HRP-conjugated secondary antibody. True Blue chromogenic substrate (KPL) was used to count the number of foci. Plaque reduction assay. Each viral isolate was titrated by the limiting dilution method and the viral titre calculated by the Reed-Muench formula. The viral titre was also calculated as plaque forming units (PFU) on six-well flat-bottomed plates (Corning, Corning, NY, USA). Neutralizing assays were carried out in 6 wells plates using MDCK cells (5610 5 cells/well). Two dilutions, 1-0.1 mg/ml, of IgG PN-SIA28 were preincubated for 1 hour at 37uC with 100 TCID 50 each of H1N1 or H3N2 virus. Following this incubation, 1 ml each of virus-antibody mix was added on MDCK monolayer and the plate was incubated 1 hour at 37uC in 5% CO 2 . After this incubation, the medium was removed and the monolayer washed twice with PBS. Two ml of MEM-agarose 0.8% supplemented with penicillin (50 mg/ml) (Gibco Invitrogen, Carlsbad, CA, USA), streptomycin (100 mg/ml) (Gibco Invitrogen, Carlsbad, CA, USA), L-glutamine (2 mM) (Gibco Invitrogen, Carlsbad, CA, USA) and trypsin (2 mg/ml) (Roche Applied Sciences) were gently added to each well and the plates were incubated 48 hours at 37uC in 5% CO2. After this incubation the agarose medium was removed from each well and 1 ml of 70% methanol-crystal violet 1% (w/v) was added to each well at room temperature. Finally, the wells were washed with tap water and dried. An infection control without antibody was added as well as a negative control with anti-HCV/E2 e137 mAb. The neutralization was determined counting the PFU reduction in presence of antibodies in comparison with the infection control. A phagemidic 12-mer peptide library (Ph.D.12 TM Phage Display Peptide Library, New England Biolabs Inc., Boston, MA, USA) was screened against PN-SIA 28 Fab fragment. Briefly, 300 ng of PN-SIA 28 were coated on four wells (96-wells plate, COSTAR), and the peptide library was amplified transforming electrocompetent E. coli XL1 Blue. 70 ml of the obtained phage preparation was incubated with PN-SIA 28 for 1 hour at 37uC and subsequently the unbound phages were removed by washing with PBS 1X-Tween20 using progressive Tween20 concentration starting from 0.1% to 0.5%. The phage that bound PN-SIA 28 was then eluted using low pH and, once neutralized, used to infect E. coli XL1 Blue for a subsequent ELISA screening on 100 ng/well coated PN-SIA 28. Positive clones were then sequenced using the kit-provided primers (New England Biolabs Inc., Boston, MA, USA). Positive clones were then sequenced. An in silico analysis was performed, using Mimox and Pepitope servers, comparing the selected peptides to the available crystal structure of A/PR/8/34 and A/South Carolina/1918 haemagglutinins (1RU7.pdb and 1RD8.pdb). The experiment was performed on 90% confluent MDCK cells growing in T25 flasks (Nunc, Rochester, NY, USA) in MEM supplemented with 10% FBS. PN-SIA28 as well as e137 mock control, were diluted in 1.5 mL MEM supplemented with TPCK trypsin (2 mg/ml) to obtain final concentrations of 2 mg/ml, 10 mg/ml and 20 mg/ml of antibody. 100 TCID 50 of A/PR/8/34 were prepared in 1.5 mL of the same medium. The two solutions, containing antibodies and virus, were mixed in order to obtain a final concentration of 1 mg/ml, 5 mg/ml and 10 mg/ml for the mAbs in a final volume of 3 mL. The mixes were then incubated 1 h at 37uC. An infection positive control (virus without PN-SIA28) was included as well as non-infected cells. After two washes with sterile PBS 16 (Gibco Invitrogen, Carlsbad, CA, USA), 1 ml of each neutralization mix was added to the flasks containing MDCK cells and the infection was performed for 1 h at 34uC in 5% CO 2 . After absorption, the medium was removed and the monolayer washed twice with sterile PBS. 3 mL of MEM supplemented with trypsin (2 mg/ml) was added to the infection positive control and to non-infected cells. Antibody PN-SIA28 and e137 were added to the previously treated infected cells, maintaining the concentrations used during the infection step. Cells were incubated at 34uC in 5% CO 2 and regularly checked for 48 hours for the presence of cytopathic effect (CPE) by comparing the positive infection control to the treated infected cells. The supernatant was then centrifuged (2000 rcf for 10 minutes), collected and stored at 280uC. All the viral stocks were titrated and used to infect new cell preparations, increasing, when it was possible, the concentration of PN-SIA28. After ten passages, the cells of the infection positive control and mock control were compared with infected cells under selective pressure of PN-SIA28. Once a strong cytopathic effect was evident in the positive and mock controls, the presence or absence of CPE in the PN-SIA28 treated infected cells was evaluated. All the supernatants were collected, centrifuged, stored and used to perform a full length DNA sequence of influenza genomic fragment 4 coding for viral HA. A/PR/8/34 (H1N1) HA was amplified as previously described [19, 20] using the following PCR-oligonucleotides: APR834_s: 59-CACCATGAAGGCAAACCTACTGGTCCTGTTATGTG-39; APR834_as: 59-TCAGATGCATATTCTGCACTGCAAAGAT-CCATTAGA-39. A/Aichi/2/1968 (H3N2) HA was amplified using the following PCR-oligonucleotides: Aichi/2/68H3N2f: 59-CACCATGAAGACCATCATTGCTTTG-39; Aichi/2/68H3-N2r: 59-TCAAATGCAAATGTTGCACCTAATG-39. The PCR products were cloned into the pcDNA 3.1D/V5-His-TOPO vector (Invitrogen, Carlsbad, CA, USA). Subsequently, alanine mutants for H1N1 and H3N2 HA were generated using Gene Tailor Site-Directed Mutagenesis System (Invitrogen, Carlsbad, CA, USA). A total of 20 HA mutants were generated for H1N1 and 7 for H3N2 (Table S1 ). The binding activity of PN-SIA28 was assayed using full-length wild type and mutants HAs cloned as described above. In brief, 1610 6 human epithelial kidney (HEK) 293T cells (ATCC CRL-1573) were transfected in 6 wells plate (Corning, Corning, NY, USA) with 4 mg of pcDNA 3.1D/V5-His-TOPO vector containing the HA nucleotide sequences. After centrifugation and fixation with paraformaldehyde 4%, the transfected cells were incubated for 30 minutes at room temperature with PN-SIA28 or a conformational control for H1N1 (RB62) or a control for H3N2 (12D1, kindly provided by P.Palese) at 1 mg/ml and 10 mg/ml. Additionally, the isotype control, e137 (1 and 10 mg/ml) was introduced as well as untransfected cells and a mouse anti-H1 subtype monoclonal antibody directed against a linear epitope (anti-influenza A hemagglutinin [12D1102] GeneTex Inc., Irvine, CA, USA) to evaluate the transfection efficiency for each HA. The cells were then washed and incubated for 30 minutes at room temperature with FITC-conjugated anti-human (Sigma Aldrich) or anti-mouse (Argene, Shirley, NY, USA) monoclonal antibodies. Afterwards, the cells were washed and analysed by FACS. The FACS data were analyzed using the software Weasel w 2.5 (Waler+Eliza Hall, Institute of Medical Research, Parkville Victoria, Australia). The binding of PN-SIA28 to the different HA-mutants was expressed as a binding percentage compared to wild-type. For sequences analysis the following software packages were used: SeqScape (Applied Biosystems), ClustalX (Toby Gibson), Bio Edit (Tom Hall, Ibis Therapeutics), and Treeview (GubuSoft).

Tissue Tropism and Target Cells of NSs-Deleted Rift Valley Fever Virus in Live Immunodeficient Mice

BACKGROUND: Rift Valley fever virus (RVFV) causes disease in livestock and humans. It can be transmitted by mosquitoes, inhalation or physical contact with the body fluids of infected animals. Severe clinical cases are characterized by acute hepatitis with hemorrhage, meningoencephalitis and/or retinitis. The dynamics of RVFV infection and the cell types infected in vivo are poorly understood. METHODOLOGY/PRINCIPAL FINDINGS: RVFV strains expressing humanized Renilla luciferase (hRLuc) or green fluorescent protein (GFP) were generated and inoculated to susceptible Ifnar1-deficient mice. We investigated the tissue tropism in these mice and the nature of the target cells in vivo using whole-organ imaging and flow cytometry. After intraperitoneal inoculation, hRLuc signal was observed primarily in the thymus, spleen and liver. Macrophages infiltrating various tissues, in particular the adipose tissue surrounding the pancreas also expressed the virus. The liver rapidly turned into the major luminescent organ and the mice succumbed to severe hepatitis. The brain remained weakly luminescent throughout infection. FACS analysis in RVFV-GFP-infected mice showed that the macrophages, dendritic cells and granulocytes were main target cells for RVFV. The crucial role of cells of the monocyte/macrophage/dendritic lineage during RVFV infection was confirmed by the slower viral dissemination, decrease in RVFV titers in blood, and prolonged survival of macrophage- and dendritic cell-depleted mice following treatment with clodronate liposomes. Upon dermal and nasal inoculations, the viral dissemination was primarily observed in the lymph node draining the injected ear and in the lungs respectively, with a significant increase in survival time. CONCLUSIONS/SIGNIFICANCE: These findings reveal the high levels of phagocytic cells harboring RVFV during viral infection in Ifnar1-deficient mice. They demonstrate that bioluminescent and fluorescent viruses can shed new light into the pathogenesis of RVFV infection.

Rift Valley fever virus (RVFV) is an arthropod-borne member of the Bunyaviridae family, genus Phlebovirus that causes recurrent outbreaks affecting humans and animals. The virus can be transmitted by Aedes and Culex mosquitoes [1] , although it can also be transmitted by inhalation or physical contact with the body fluids from infected animals [2, 3] . Identified in the 1930s in Kenya, RVFV has spread during recent years to most sub-Saharan African countries, in Egypt and in the Arabian Peninsula, and in the Indian Ocean islands of Grande Comore and Mayotte [4, 5, 6] . In humans, RVFV infections are generally either asymptomatic or characterized by a feverish syndrome without any severe sequelae. However, a small percentage of patients exhibit complications, characterized by acute hepatitis with hemorrhage, meningoencephalitis and/or retinitis [7, 8, 9, 10] . A relationship has been demonstrated between high viral load in blood and death of the patient [11, 12] . RVFV infects domestic ruminants, including sheep, cattle, goats, and camels. It is responsible for massive abortion events in pregnant ruminants and high mortality in lambs and calves. High viremia associated with hepatic necrosis and increase of liver enzymes are hallmarks of severe acute lethal infection in ruminants [13, 14] . Encephalomyelitis has been described in calves [15] . Laboratory rodents such as mice are also highly susceptible to RVFV infection. In outbred Swiss mice, the survival time was inversely proportional to the logarithm of the viral dose inoculated via the intravenous route [16] . Depending on their genotype, males from various inbred strains of mice inoculated by the peritoneal route with 10 2 PFU of the virulent Egyptian ZH548 strain die between 4 to 10 days after inoculation, illustrating natural variation in susceptibility of the host to RVF [17] . The main damages of mouse infection with RVFV can be observed early in the liver, with extensive apoptosis of hepatocytes, accompanied in the blood by a peak in liver enzymes, along with increased bilirubin levels [18, 19, 20] . It has been recently shown that mice that survive hepatitis develop later infection of the brain, and eventually die from meningoencephalitis [21] . Interestingly, a diverse set of cell types from a number of tissues was found to contain RVFV antigens, including mononuclear phagocytes, but also cardiac myofibers, pancreatic islet cells, and adrenal medullary cells [21] . These data showed that RVFV exhibits a large tropism for a variety of tissues and individual cell types. Quantitative real-time PCR have also been used to study the kinetics of RVFV infection in the blood and organs of infected mice [22] . High amounts of RVFV RNA were found in blood, liver and brain samples shortly after infection with the highest viral RNA levels in the liver. We hypothesized that in vivo imaging might be an alternative method to assess viral replication using a recombinant RVFV carrying a reporter gene that allows the monitoring of viral expression in live animal. RVFV has a tripartite negative-sense, single-stranded RNA genome with large (L), medium (M) and small (S) segments. The L segment encodes the viral RNA-dependent RNA polymerase, the M segment the two virion glycoproteins (G N and G C ) and the NSm nonstructural proteins, and the S segment the N nucleoprotein and the NSs nonstructural protein. Reverse genetic systems have been successfully developed for the recovery of recombinant RVFV (reviewed in [23] ). These rescue systems rely on transfection with plasmids expressing the three viral RNAs, and the N nucleoprotein and L RNA-dependent RNA polymerase, which are required for the packaging and replication of the viral RNAs. The RVFV RNA genome segments are expressed under the control of either the cellular DNA-dependent RNA polymerase I promoter [24, 25, 26] or the bacteriophage T7 promoter in cells that constitutively express T7 RNA polymerase [24] . Such rescue systems have been used to produce recombinant RVFV, including various mutants that lack the NSs, NSm genes, or carry specific mutations [26, 27, 28, 29, 30] . Viral strains that express reporter genes were also generated [24, 31, 32] . Importantly, in these recombinant viruses, the reporter gene activity directly reflects the extent of both viral transcription and replication. In this study, we aimed to detect and quantify viral replication in living animals using two recombinant RVFV strains expressing either humanized version of the luciferase gene of Renilla reniformis (hRLuc) or enhanced green fluorescent protein (GFP) gene of Aequora victoria. Both RVFV viruses lack a functional NSs gene, which is a main factor of virulence in mice [33] , and are therefore avirulent in immunocompetent mice. However, in mice that are nonresponsive to type I IFN, the virus expressing either hRLuc or GFP caused lethality within 3 days, in agreement with previous data for the NSs-deficient Clone 13 [24] . In these mice, virus infection could be tracked by luciferase imaging in live animals and by the detection of GFP-positive cells from infected animals, by use of flow cytometry. We observed qualitative and quantitative differences in the in vivo tropism of RVFV in mice and show previously unsuspected sites of virus replication and modes of virus spread. Animals were housed in the Institut Pasteur animal facilities accredited by the French Ministry of Agriculture to perform experiments on live mice, in appliance of the French and European regulations on care and protection of the Laboratory Animals (accreditation number B 75 15-01 and B 75 15-07). The veterinary staff of the Institut Pasteur animal facility approved protocols. Protocols were performed in compliance with the NIH Animal Welfare Insurance #A5476-01 issued on 02/07/2007. Inbred 129S2/SvPas mice with knockout at the interferon a and b receptor 1 locus (Ifnar1 2/2 ) and control mice (Ifnar1 +/+ ) were bred at the Institut Pasteur [34] . Vero E6 cells were grown in DMEM supplemented with 10% FCS. BHK21/T7 cells [35] were grown in MEM supplemented with 5% FCS and tryptose phosphate broth powder (Sigma-Aldrich, Gillingham, UK). The cell culture media were supplemented with 10 IU/ml of penicillin and 10 mg/ ml of streptomycin. Stocks of the virulent RVFV Egyptian ZH548 strain were produced under biosafety level 3 (BSL3) conditions. Plasmids pPol I-LZH, pPol I-MZH, and pPol I-SZH carrying the L, M and S segments of ZH548, respectively, were cloned in the plasmid pRF108, [24, 36] . The plasmid pPol I-SZHDNSs, derived from pPol I-SZH, carries two BbsI cloning sites in place of NSs [24] . The humanized Renilla reniformis luciferase sequence from phRL-SV40 (Promega, Charbonnières-les-Bains, France) was inserted in pPol I-SZHDNSs to give pPol I-SZHDNSs-hRLuc. The structure of pPol I-SZHDNSs-hRLuc plasmid was confirmed by sequence analysis. Recombinant rZHDNSs-GFP [24] and rZHDNSs-hRLuc RVFV stocks were produced under BSL3 conditions. Approximately 5610 5 BHK21/T7 cells were seeded in triplicate in 35 mm culture dishes. The following day, they were combined with FuGENEH6 transfection reagent (Roche Applied Science, Indianapolis, IN) and 0.5 mg each of pTM1-L and pTM1-N [25] , and 1 mg each of pPol I-LZH, pPol I-MZH, and either pPol I-SZHDNSs-GFP or pPol I-SZHDNSs-hRLuc in OptiMEM (Gibco, invitrogen, Carslbad, CA). One day later, the medium was renewed. Five days later, the supernatant containing the rescued virus was collected and stored at 280uC. To produce Rift Valley fever, caused by a member of the Bunyaviridae family, has spread during recent years to most sub-Saharan African countries, in Egypt and in the Arabian peninsula. The virus can be transmitted by insect vectors or by direct contacts with infectious tissues. The analysis of virus replication and dissemination in laboratory animals has been hampered by the need to euthanize sufficient numbers of animals and to assay appropriate organs at various time points after infection to evaluate the viral replication. By following the bioluminescence and fluorescence of Rift Valley fever viruses expressing light reporters, we were able to track the real-time dissemination of the viruses in live immunodeficient mice. We showed that the first infected organs were the thymus, spleen and liver, but the liver rapidly became the main location of viral replication. Phagocytes also appeared as important targets, and their systemic depletion by use of clodronate liposomes decreased the number of viruses in the blood, delayed the viral dissemination and prolonged the survival of the infected mice. viral stocks of rZHDNSs-GFP and rZHDNSs-hRLuc, Vero E6 cells were infected with the rescued virus at a MOI of 0.001 and 0.01, respectively. At 72 h post-infection, the supernatant was collected and the viral suspension was titered. Vero E6 cells, infected with serial dilutions of viral suspension, were incubated under an overlay of DMEM supplemented with 2% FCS, antibiotics and 1% agarose. Four days later, the plates were stained with 0.2% crystal violet in 10% formaldehyde, 20% ethanol and the lytic plaques were counted. To test the luciferase expression within cells after infection with rZHDNSs-hRLuc, Vero E6 cells were infected using a MOI of 0.3 and 3. Next, every 2 h, for 12 h, luciferase activity was measured in triplicates by Renilla Luciferase Assay System (Promega, Madison, WI). To check the stability of luciferase expression through passages, Vero E6 cells were infected using an MOI of 3 and the luciferase activity was measured in triplicates at 8 h postinfection. To test the GFP expression, Vero E6 cells were infected with rZHDNSs-GFP using a MOI of 1. At 15 h post-infection, the cells were fixed for 30 min at room temperature with 4% paraformaldehyde in PBS, permeabilized for 10 min with 0.5% Triton X100 in PBS and incubated for 30 min at room temperature in 5% bovine serum albumin (BSA) in PBS. The cells were next incubated for 30 min at 37uC with a mouse anti-N antibody diluted in 5% BSA in PBS (dilution 1:800), washed with 5% BSA in PBS and incubated for 25 min at 37uC with the secondary antibody Alexa Fluor 555 goat anti-mouse (Invitrogen, Paisley, UK) diluted in 5% BSA in PBS (dilution 1:1200). Finally, the cells were washed with 5% BSA in PBS and then in water. The slides were mounted with Fluoromont-G (SouthernBiotech, Birmingham, AL). The cells were observed under an Axioplan 2 Imaging microscope (Zeiss, Le Pecq, France) using excitation and emission filter allowing simultaneous detection of GFP and Alexa Fluor 555. One week prior to infection, five to six week-old mice were transferred in BSL3 isolators to allow acclimatization. After this period, they were inoculated intraperitoneally (i.p.), intradermally (i.d.) or intranasally with 10 4 PFU ZH548, rZHDNSs-hRLuc or rZHDNSs-GFP RVFV in DMEM supplemented with 2% FCS, and antibiotics. For i.p. infection, mice were inoculated with 100 mL of viral suspension. For i.d. and intranasal infections, mice were first anesthetized with ketamine (150 mg/kg) and xylazine (10 mg/kg) administered i.p., then inoculated with either 15 to 30 mL or 15 mL of viral suspension into the ear (i.d.) or intranasally, respectively. Mortality was recorded at least twice a day from day 1 to 4 post-infection and once a day after day 5 postinfection until the end of the observation period. Animals were observed for a maximum of 14 days. To improve bioluminescence imaging, hairs were removed [37] . We observed the mice to detect clinical signs due to the infection prior to imaging. Mice that exhibited no severe clinical signs were anesthetized and injected i.p. with 100 mL h-coelenterazine (1 mg/ml). The h-coelenterazine stock solution provided by Nanolight Technology (Pinetop, AZ) was solubilized in ethanol-propylene glycol solution (1:1) at 10 mg/ml. This solution was diluted in PBS (1:9) just before imaging. The h-coelenterazinetreated mice were immediately placed in a hermetically sealed light-tight-transparent chamber (TEM Sega, Lormont, France) equipped with two HEPA filters. One HEPA filter was connected to an air pump, thus allowing air renewal during the imaging. The bio containment chamber allowed simultaneous imaging of 6 mice. The mice were imaged 15 and 20 min after h-coelenterazine injection for the whole body and the thorax, respectively [38] . Imaging was performed with a Xenogen's IVIS 100 system, including a cooled charge-coupled device (CCD) camera [39, 40] . Integration periods ranged from 0.5 to 120 s depending on the amounts of light emitted at various infection sites. Images were obtained using Living ImageH 3.1 software (Xenogen, Alameda, CA). Specific regions of interest (ROI) on the images were defined without overlay, using the anatomic location of the different organs and their visual observation through the skin when possible. For each imaging session, a mock-infected mouse was used as a negative control. A signal was considered significant if its intensity in infected mice was at least twofold higher than the background luminescence in the mock-infected mouse. After the last imaging time point, mice were euthanized. For ex vivo imaging, selected mice were euthanized at different times after infection for harvest of the following organs: liver, spleen, thymus, lung, kidney, stomach, small and large intestine, heart, ovary and uterus, testis, epididymis, seminal vesicles and preputial glands. The organs were placed in 6 (intestine) or 2 ml (all other organs) of PBS. Before imaging, 1 mL/ml h-coelenterazine 5 mM in ethanol and propylene glycol (1:1) was added [41] . Imaging was performed in a hermetically sealed chamber to avoid light. Images were acquired 10 min after the addition of hcoelenterazine. Integration period ranged from 0.5 to 60 s depending on the amounts of light emitted from various organs. RNA was extracted using Trizol LS reagent (Invitrogen, Carslbad, CA) and suspended in RNase free water. RNA was quantified using Nanodrop 3300 (Thermo Scientific, Courtaboeuf, France). The M segment of RVFV was amplified with primers 59-CATGGATTGGTTGTCCGATCA-39 and 59-TGAGTGTAA-TCTCGGTGGAAGGA-39. Quantitative RT-PCR assays were performed using StepOne Plus Real-Time PCR System (Applied Biosystem, Courtaboeuf, France) in 96-well plates. Reverse transcription using MultiScribe Reverse Transcriptase (Applied Biosystem) at 48uC for 30 min was performed followed by a standard amplification program. The size of the amplification product was 108 pb. A standard curve was generated using duplicates of 10-fold serial dilutions of RNA of the M segment ranging from 10 9 to 10 2 copies. Quantification of viral RNA was done by comparison of the threshold cycle (Ct) values of the samples to the standards. Histopathological and immunohistochemical analysis of wildtype mice infected with 10 4 PFU ZH548 RVFV was performed 3 to 5 days after i.p. inoculation. rZHDNSs-hRLuc-infected Ifnar1 2/2 mice were euthanized at 8, 16 and 34 h after i.p. inoculation. For each time point, a complete post-mortem examination was carried out. The lung, brain, kidneys, spleen, liver, pancreas, thymus, testis, uterus and ovaries were removed and immediately fixed for one week in 10% neutral buffered formalin. Samples from each organ were embedded in paraffin and five-micrometer sections were then cut and stained with hematoxylin and eosin (HE). The histological characterization of lesions was completed by an immunohistochemical detection of the RVFV using mouse antibodies against the RVFV (dilution 1:100) visualized with the Histofine Simple Stain MAX-PO kit (Histofine Biosciences inc, Cambridge, UK). Eleven Ifnar1-deficient 129S2/SvPas mice were either infected i.p. with 10 4 PFU rZHDNSs-GFP RVFV (N = 6) or mock-infected (N = 5). Twenty-four hours later, the spleen was harvested. Erythrocytes were lysed using NH 4 Cl (9 g/L) buffer. The rat anti-mouse CD16/CD32, clone 2.4G2 antibody (BD Pharmingen, San José, CA) was used to block non-antigen-specific binding of immunoglobulins to Fc-receptors. Cells were stained using a combination of the following antibodies: (i) PE-conjugated rat anti-mouse NKp46/CD335 (BD Pharmingen). (ii) PerCP-Cy5.5-conjugated hamster antimouse CD3 (BD Pharmingen). (iii) APC-conjugated rat anti-CD19 (BD Pharmingen). (iv) Pacific Blue-conjugated rat antimouse CD11b/Mac-1 (eBioscience, San Diego, CA). (v) APCconjugated hamster anti-mouse CD11c/Itgax (BD Pharmingen). (vi) Alexa Fluor 700-conjugated rat anti-mouse MHC Class II (I-A/I-E) (eBioscience). (vii) PE-conjugated rat antimouse Ly6G/Gr-1 (BD Pharmingen). (viii) Biotin-conjugated anti-mouse CD115/c-Fms (eBioscience) with Streptavidin-PerCP-Cy5.5 (BD Pharmingen) as second-step reagent. All staining procedures were conducted on ice. Then, the cells were fixed with 4% formaldehyde. Fluorescence was measured using a FACSAria II flow cytometer (BD Biosciences, San Jose, CA), and data analysis was performed using CellQuest (BD Biosciences) and FlowJo (Ashland, OR) softwares. Dead cells were visualized using the Fixable Aqua Dead Cell Stain kit (Invitrogen, Carlsbad, CA). Fluorescence compensation settings for multicolor flow cytometric analysis were optimized based on single-stained polystyrene microparticles (Comp-Beads, BD Pharmingen). Clodronate (Cl2MBP; dichloromethylene-biphosphonate)-loaded liposomes (CLL) were used to deplete phagocytic cells [42, 43] . Clodronate was a gift of Roche Diagnostics GmbH, (Mannheim, Germany). It was encapsulated in liposomes as described earlier [42, 43] . Mice were injected i.p. with 300 mL and i.v. with 200 mL CLL. Control mice were treated i.p. and i.v. with PBS-loaded liposomes. Twenty-four hours later, single-cell suspensions were prepared from blood and spleen. FcR blocking reagent mouse (Miltenyi Biotec, Bergisch Gladbach, Germany) was used to block non-antigen-specific binding of immunoglobulins to Fc-receptors. Cells were stained using combination of the following antibodies: FITC-conjugated rat anti-mouse CD11b (BD Pharmingen), PEconjugated rat anti-mouse CD115 (eBioscience.com), APCconjugated rat anti-mouse F4/80 (eBioscience.com), PE-conjugated hamster anti-mouse CD11c (BD Pharmingen) and Pacific Blueconjugated rat anti-mouse Ly6G/Gr-1 (eBioscience.com). All staining procedures were conducted on ice. Fluorescence data were obtained and analyzed using MACSQuant Analyzer and MACSQuantify software (Miltenyi Biotec). Challenge with 10 4 PFU rZHDNS-hRLuc was performed by injection into the ear, 24 h after liposome treatment. The survival curves were compared using the logrank test. The bioluminescence signals and blood plasma viral loads were analyzed with the nonparametric Mann-Whitney test. All tests were performed using the StatView 5.0 software (SAS Institute Inc, Cary, NC). The generation of a recombinant RVFV expressing a green fluorescent protein (GFP), rZHDNSs-GFP, has previously been described [24] . Our previous attempts to generate a recombinant RVFV expressing a humanized firefly luciferase (hFLuc) have been confounded by genetic instability and the rapid emergence of mutants with deletions [24] . Therefore, we generated rZHDNSs-hRLuc RVFV that carries a humanized Renilla luciferase (hRLuc) gene using Pol I based plasmids as previously described [24] . The rescued rZHDNSs-hRLuc was amplified in Vero E6 cells and stocks produced. The titer reached 8610 7 PFU/ml. The plaques formed by rZHDNSs-hRLuc were fuzzy with a faint staining inside, as those obtained with the rZHDNSs virus that carries a deletion of the NSs gene [24] . To test the luciferase expression within the infected cells, Vero E6 cells were infected with rZHDNSs-hRLuc at a MOI of either 0.3 or 3, and lysed every 2 hours for 12 h. Luciferase activity was measured using coelenterazine, a specific substrate of Renilla luciferase, and found to be expressed at significant levels from 2 h post-infection onwards, while uninfected Vero E6 cells showed no luciferase activity. The luciferase activity increased with time and was dependent on the MOI (data not shown). This is consistent with previous reports [31, 32] . To check the stability of the recombinant virus in vitro, rZHDNSs-hRLuc was passaged on Vero E6 cells and, at each passage, we measured the viral titer in the supernatant at 72 h post-infection and the luciferase activity within the infected cells at 8 h post-infection with a MOI of 3. Both the viral titer and the luciferase activity remained stable over at least 8 passages, varying from 10 7 to 10 8 PFU/ml and from 10 7 to 10 8 raw light units (RLU)/s per 3610 5 cells, respectively. To test the virulence of the recombinant virus, wild-type 129S2/SvPas mice (N = 5) and 129S2/SvPas mice deficient for IFN-a/b receptor subunit 1 (Ifnar1 2/2 ) (N = 10) were infected i.p. with 10 4 PFU of rZHDNSs-hRLuc. All wild-type mice survived the infection for 13 days with no signs of disease, as seen following infection with rZHDNSs in which the NSs gene is totally deleted [24] or Clone 13, a natural isolate that lacks 69% of the NSs open reading frame [33] . In contrast, all rZHDNSs-hRLuc infected Ifnar1 2/2 mice died within 45 h (Figure 1 ) from severe hepatitis with no signs of neurological disorder. Infections of Ifnar1 2/2 mice with rZHDNSs or with Clone 13 gave similar results ( [24] and data not shown). To evaluate the stability of the recombinant viruses in live animals, total RNAs were extracted from the liver of rZHDNSs-hRLuc-or rZHDNSs-GFP-infected Ifnar1 2/2 mice at 34 h postinfection and RT-PCR assays were performed using primer pairs flanking the hRLuc or GFP reporter gene. The amplification product sizes were those expected from the structure of pPolI-SZHDNSs-hRLuc and pPolI-SZHDNSs-GFP plasmids (data not shown). No amplification products with smaller sizes were observed, suggesting that the recombinant viruses maintained their own genomic stability not only in cultured cells, but also during in vivo infection. Furthermore, to examine the reporter expression stability after in vivo infection, the recombinant rZHDNSs-GFP was harvested from the liver of an infected Ifnar1 2/2 mouse at 34 h post-infection and then used to infect Vero E6 cells at a MOI of 1. The percentage of cells positive for the N viral protein that were also GFP-positive was measured. The percentage of N-positive, GFP-positive cells was almost identical to that of the initial viral stock (84%61.90% vs. 85%67.34%). Altogether, these results suggest that the recombinant viruses were stable for the time of infection in live mice. To visualize the spread of the virus, Ifnar1 2/2 mice were infected i.p. with 10 4 PFU of rZHDNSs-hRLuc. At 8, 16 and 34 h postinfection, h-coelenterazine was injected i.p. This route of hcoelenterazine administration was preferred to tail-vein injection due to slower kinetics of light production, as previously reported [39] . Mice were observed with real-time in vivo imaging 15 and 20 min after the injection of h-coelenterazine for the whole body and the thorax, respectively. At 8 h post-infection, luminescence was readily detected. Short integration periods (15 s) were sufficient to acquire a significant signal. We observed strong signals between the forelegs in the thoracic cavity, and below the xiphoid cartilage in the abdominal cavity, respectively (Figure 2A) . Imaging of the left profile showed an additional signal in the spleen ( Figure 2B ). In some experiments, animals were euthanized for ex vivo imaging, the organs of the thorax and abdomen were harvested, and the individual organs were imaged ( Figure 2D ). In the thorax, the greatest signal originated from the thymus, whereas the signal from the lungs was only slightly above background. In the abdomen, the pancreas was the most luminescent organ. The spleen and the liver also emitted significant luminescence. On average, a ten-fold higher luminescence signal was observed in the harvested pancreas compared to the liver. This suggests that the liver, the critical target organ of the disease, was not among the main replication sites for RVFV at this early stage of infection. At 16 h post-infection, the whole body luminescence was higher than at 8 h post-infection (Figure 2A) . The signal spread out the abdominal cavity. The high intensity of luminescence in the abdominal cavity precluded detection in other locations unless integration was limited to the thorax ( Figure 2C ). Dissection and ex vivo imaging showed a gradual increase of luminescence in the pancreas and in the liver. Additional sources of luminescence were the intestine mesentery, kidneys, ovaries and uterus in females, the seminal vesicles, preputial glands, epididymis and testis in males ( Figure 2D , and data not shown). The intensity of these signals was quite similar to that measured in the spleen and in the liver (data not shown). At 34 h post-infection, the intensity of the signal led to saturation of the camera using a 0.5 s integration period, thus preventing identification of individual organs (Figure 2A) . Ex vivo imaging revealed that the highest signal was in the liver. Other organs with intense luminescence were the spleen, intestine mesentery and pancreas (data not shown). Quantification of the luciferase expression in living mice during the time course of infection showed that the luminescence signal originated from the thymus remained constant from 8 h postinfection onwards, whereas luminescence profiles were increased in the liver and pancreas, suggesting a progressive increase of viral replication in these organs (Figure 3 ). To determine whether there was a correlation between the luminescence detected by the camera and the amount of virus genomes in the tissues, rZHDNSs-hRLuc-infected Ifnar1 2/2 living mice were subjected to imaging. Next, the animals were euthanized and the organs harvested and imaged. Total RNAs were extracted from the organs and RVFV RNA copy numbers were measured by quantitative real time RT-PCR. We observed a highly significant correlation between the luminescence emitted by the pancreas in living mice and RVFV RNA copy number ( Figure 4A ). Similarly, luminescence intensity significantly correlated with the RVFV RNA copy number in the harvested pancreas, spleen and liver ( Figure 4B-D) . The ability of h-coelenterazine to cross the blood-brain barrier is unknown. To determine whether rZHDNSs-hRLuc can infect the brain, we dissected and soaked the brain in an hcoelenterazine solution and imaged ( Figure 2D ). At 8 h postinfection, the luminescence intensity was ten-fold higher in the brain from infected mice compared to control (10 4 photons/ second/cm 2 /steradian [p/sec/cm 2 /sr] vs. 10 3 p/sec/cm 2 /sr), showing that the RVFV replicated in the brain at an early stage. Light emission increased through 16 h and 34 h post-infection to reach 7610 5 and 7610 6 p/sec/cm 2 /sr, respectively. Importantly, the intensity of luminescence in the brain was ten-to hundred-fold lower compared to the intensities in the thymus, pancreas, spleen, liver, and intestine mesentery, suggesting that the viral load was lower in the brain than in the thoracic and abdominal organs. RVFV can be transmitted through injection of infectious saliva from mosquito into the dermis or direct inhalation from body fluids, such as blood of infected animals [2, 3] . To approximate these two natural routes of infection and to monitor their effects, we compared the light production after intraperitoneal, intradermal or intranasal inoculation of 10 4 PFU rZHDNSs-hRLuc into Ifnar1 2/2 mice. Following intradermal inoculation of the ear pinna (N = 5), luminescence was first visible in the neck on the side of the injected ear, and in the abdominal cavity at 24 h postinfection ( Figure 5A ). Histologic analysis established the source of the light in the neck; the neck signal came from the lymph nodes draining the injected ear (data not shown). At 40 h post-infection, organs in the abdominal cavity, including the pancreas and the liver were highly luminescent. All mice succumbed to infection by 69 h post-infection, a survival time significantly longer than after i.p. inoculation (P,0.025). Luminescent virus inoculated intranasally was already detected 24 h post-infection in the abdomen. This mode of inoculation caused an interstitial pneumonia that initiated as a distinctive luminescence signal in the lungs from 48 h post-infection onwards ( Figure 5B ). Mice infected intranasally (N = 5) survived significantly longer than after i.p. inoculation (P,0.0047); all were dead by 69 h post-infection. To clarify the identity of the RVFV target cells, we carried out histopathological analysis in RVFV-hRLuc-infected Ifnar1 2/2 No histological lesions were detected in the other organs. In the liver, lung and spleen, the lesions were similar to those previously reported in RVFV-infected wild-type mice [18, 19, 20, 21] . Diffuse apoptosis of lymphoid cells have been previously reported in areas with or without RVFV antigen in the thymus of infected BALB/c [21] . Accordingly, we identified the thymus as one of the major targets of RVFV by bioluminescence. However, the pancreas and reproductive organs were also luminescent although none of these organs are known as tissue targets of RVFV. Therefore, to identify cell types that support RVFV replication in these organs, we studied tissue samples by histology and immunohistochemistry with antibodies against the RVFV. In the pancreas, no histological lesions were found in the exocrine or endocrine components ( Figure 6A ). However, a multifocal inflammatory lesion was observed in the mesentery around pancreatic acini (peritonitis), characterized by necrosis of adipocytes associated with infiltration of macrophages and neutrophils ( Figure 6A-C) . Viral antigens were present only in the cytoplasm of macrophages ( Figure 6D ) and, more rarely, in neutrophils ( Figure 6E ), confirming that the virus did not target the pancreatic exocrine or endocrine cells but macrophages. Similarly, in the ovaries, viral nucleocapsid-positive macrophages were seen in the stroma ( Figure 6F ). Thus macrophages appeared as important cell targets for the replication of RVFV-hRLuc in Ifnar1-deficient mice. To examine whether macrophages are also cell targets for the replication of virulent RVFV in wild-type mice, histopathological and immunohistochemical analysis was performed in wildtype 129S2/SvPas mice (N = 3) infected i.p. with 10 4 PFU ZH548. Post-mortem analyses were carried out once mice displayed clinical signs, i.e. three to five days after the inoculation. Histopathological analysis of the pancreas and its mesentery revealed no peritonitis. However, numerous macrophages containing intracytoplasmic viral antigens were observed in the sinus of the pancreaticoduodenal lymph node (Figure 6, G and H) . These macrophages occasionally displayed a hyperbasophilic and condensed nucleus, a morphological change that is characteristic for irreversible cell injury (Figure 6 , I). Collectively, these results confirmed that macrophages are important cell targets of the RVFV in the mouse. To further dissect target cells of RVFV replication in Ifnar1deficient mice, we used the recombinant virus rZHDNSs-GFP that carries GFP in place of the NSs gene. We have shown previously that cells infected in vitro with rZHDNSs-GFP are fluorescent upon excitation at 488 nm [24] . Ifnar1-deficient mice were either infected i.p. with 10 4 PFU rZHDNSs-GFP (N = 6) or mocktreated (N = 5). At 24 h post-infection, the spleen was dissected and single-cell suspensions were analyzed by flow cytometry for GFP expression. At this time point, 0.54% (range 0.14-1.53%) of the total hematopoietic cell population of the spleen from rZHDNSs-GFP-infected mice expressed GFP whereas no GFP-positive cells were found in splenocytes after mock infection ( Figure 7A ). We examined the expression of GFP in various subsets of antigen presenting cells based on the surface expression patterns of CD45.2, CD11b, CD11c, Ly6G, CD19, CD3, NKp46, CD115 and MHCII class II by multicolor flow cytometric analysis. Among the CD11b + CD115 + Ly6G 2 (macrophages), CD11c + CD11b + MHC II + (dendritic cells) [44] and CD11b + CD11c 2 Ly6G + (granulocytes), on average 5.58% (range 2.15-8.71%), 4.5% (range 0.82-8.49%) and 1.96% (range 0.05-6.07%) cells expressed GFP, respectively ( Figure 7C [left panels], B [right panels], and C [right panels], respectively). The percentage of GFP-expressing cells within the total cell population of spleen varied from one infected mice to another, indicating that the dynamics of RVFV infection progression was not identical in all individuals. However, each of the three subsets of immune cells was infected with the same efficiency in the different mice. This is shown by the fact that the ratio of GFP-expressing macrophages, dendritic cells or granulocytes was highly correlated with the ratio of GFP-expressing cells in the total cell population from the spleen (Pearson correlation coefficient 0.97, 0.85 and 0.79, respectively) . These findings suggest a distinct pattern of susceptibility to infection by the RVFV-GFP for different immune cells in the following order: CD11b + CD115 + Ly6G 2 (macrophages).CD11c + CD11b + MHC II + (dendritic cells).CD11b + CD11c 2 Ly6G + (granulocytes). On average, less than 0.4% (range 0-1.19%) of NKp46 + CD3 2 natural killer (NK) cells were positive for GFP ( Figure 7E, left panel) . Finally, GFP fluorescence was seen on average in only 0.25% (range 0.07-0.57%) of CD19 + CD3 2 (B lymphocytes) cells and 0.20% (range 0.05-0.43%) NKp46 2 CD3 + (T lymphocytes) cells ( Figure 7D and E, right panels). Thus, at 24 h post-infection, RVFV replicated in cells of the myeloid lineage, primarily in mononuclear phagocytic cells, such as macrophages, dendritic cells and granulocytes. To study the significance of virus replication in phagocytic cells in vivo, we injected intraperitoneally (i.p.) and intravenously (i.v.) clodronate-loaded liposomes (CLL) to Ifnar1-deficient mice prior to infection with RVFV. These liposomes are widely used to deliver clodronate to phagocytic cells, especially macrophages, and the accumulation of clodronate leads to irreversible metabolic damages, which will eventually result in apoptosis [45] . As reported previously [42, 43] , the i.p. administration of CLL kills macrophages in the peritoneum and spleen of wild-type mice whereas i.v. administration affects mainly macrophages in the spleen and liver. We first analyzed the effect of CLL treatment on the phagocytic cell population of the blood and spleen of Ifnar1deficient mice 24 h after i.p. and i.v. administration. Flow cytometric analysis was performed to compare the percentage of macrophages/monocytes, dendritic cells and granulocytes in samples from mice treated with CLL (N = 3) and PBS liposomes (PBSL) (N = 3). The CLL treatment resulted in a 23-fold reduction of CD11b + CD115 + cells (monocytes) and a 6-fold reduction of CD11b + F4/80 + -expressing macrophages in the blood and spleen, respectively. In addition, CD11b + CD11c + F4/80 2 cells (dendritic cells) were decreased 9-fold in the blood. By contrast, CD11b + CD11c 2 Ly6G + cells (granulocytes) were not depleted in the blood and spleen, as previously reported [46] . Altogether this analysis showed that, 24 h after CLL treatment, blood monocytes and dendritic cells and spleen macrophages were efficiently depleted in Ifnar1-deficient mice whereas granulocytes were not affected. Next, CLL-or PBSL-treated mice were infected intradermally with 10 4 PFU rZHDNSs-hRLuc at 24 h after liposome administration. To investigate whether the clodronate treatment affected viral replication in vivo, we first observed PBSL-and CLL-administered infected mice using the imaging of whole bodies at 24 and 40 h post-infection. The profile of bioluminescence signals was similar in PBSL-and CLL-administered mice (N = 5 in each group) (data not shown). However, we observed weaker signals in CLLadministered mice. Indeed, the signals in the liver region were on average fifteen and four-fold lower at 24 and 40 h post-infection respectively in CLL-administered mice compared to PBSLtreated mice (P,0.05). We then measured the viraemia at 24 In vivo imaging studies using reporters, such as hRLuc and GFP, may provide a more complete picture of the spatiotemporal progression of a viral disease [47, 48] . In this study, we report the use of recombinant RVFV-hRLuc and RVFV-GFP strains to investigate the in vivo dynamics of RVFV infection progression in living mice and identify the virus-expressing cells. The recombinant viruses were generated by replacing the NSs gene with the reporter gene. Hence, these viruses were avirulent in immunocompetent mice when compared with wild-type virus but they were highly pathogenic in mice lacking interferon-a/b receptor, enabling to use them for pathogenesis studies in this mouse model. We were able to detect luciferase reporter expression at early stages of infection in the main known sites of viral replication, the liver, the spleen, the thymus and the brain [18, 21, 49, 50, 51] . The pancreas appeared as an unexpected site of virus replication. Using ex vivo imaging and histological examination, we primarily identified macrophages infiltrating the adipose tissue surrounding the pancreas as primarily virus-expressing cells. Similarly, RVFVexpressing macrophages were identified in the stroma of the ovary. The RVFV-GFP confirmed the importance of macrophages as specific host cells for the virus in Ifnar1 2/2 mice. It further allowed the identification of dendritic cells and granulocytes as target cells for RVFV replication. Interestingly, only a low number of B-, T-and NK-cells expressed the GFP reporter. Viral antigens have been previously detected in mononuclear phagocytic cells and dendritic cells in the lymph nodes, spleen and thymus from infected wild-type mice [21] and in macrophages in the lymph nodes from infected rats [51] . Interestingly, Smith and colleagues [21] also noticed that lymphocytes did not appear stained with the RVFV antibody in agreement with our observations. Although RVFV replication in the human macrophage-like cell line U937 [52] and in cultured peritoneal macrophages from susceptible rats [53] have been previously documented, this is, to our knowledge, the first study to evaluate the infection rates of various subsets of cells of the myeloid lineage in vivo. Because the level of fluorescent GFP directly reflects the extent of transcription and replication of the recombinant virus, we assume that the virus is highly expressed in GFP-positive cells. However, since macrophages, dendritic cells and granulocytes are able to uptake cell debris, it is possible that some of these cells are GFP-positive following phagocytosis of debris of RVFV-GFPinfected cells in vivo. Phagocytic cells function as pathogen sensors. Macrophages and neutrophils provide the first line of defense following infections. Macrophages and dendritic cells are antigen presenting cells and play a crucial role in the establishment of the adaptive immune response. Infection with RVFV-GFP showed that phagocytic cells are also target cells for RVFV. We investigated the importance of the in vivo interaction between phagocytic cells and RVFV. We treated Ifnar1-deficient 129S2/SvPas mice with CLL to deplete the population of phagocytic cells, and showed that, following intradermal infection with RVFV, the depleted mice allowed reduced RVFV replication compared to control mice, as assessed both by in vivo imaging and viral titration from blood samples. Accordingly, CLL-treated mice displayed enhanced survival time compared with control mice, indicating that phagocytic cells are involved in the pathogenesis of RVF. Altogether, our data suggest that during the initial stages of infection of Ifnar1 2/2 mice, the virus replicates inside macrophages and dendritic cells. On the other hand, since the RVFV replicates in diverse cell types in peripheral tissues, the infection may progress rapidly and lead to acute hepatitis and death. RVFV is thought to be transmitted primarily by bites of infected mosquitoes, by direct contact with infected body fluids or through airborne transmission. It has been confirmed that exposure of mice to aerosols containing RVFV is able to induce infection [54] . Following inoculation into a dermal site, RVFV-hRLuc expression was seen in the draining lymph node which became the main site of replication early after infection while the virus was still weakly detected into the abdominal cavity. Later, virus spread and caused severe hepatitis within 69 h post-infection. Following intranasal inoculation, virus replicated in the lung where it caused pneumonia within 48 h post-infection. Its dissemination to the abdominal cavity was rapid and mice succumbed at 69 h postinfection. Thus, typical routes of exposure were associated with clear differences in the spatial and temporal progression of RVFV and caused delayed death compared with i.p. inoculation. Type I interferons (IFNs) are essential elements during host antiviral defense [55] . Both recombinant RVFV strains inoculated i.p. were able to kill Ifnar1-deficient 129S2/SvPas mice within 2 days whereas wild-type 129S2/SvPas mice survived infection, indicating that a functional IFN-a/b pathway is critical for the protection of mice from fatal infection with these attenuated viruses. We showed that the recombinant viruses could replicate in known target tissues and cells of RVFV. It is not clear whether in the absence of the IFN-a/b receptor, the reporter RVFV can replicate in tissues and cells that are not normally susceptible to infection with a fully virulent RVFV in wild-type mice. Hence, we infected wild-type 129S2/SvPas with the virulent RVFV ZH548 strain and observed infected macrophages in the spleen and pancreaticoduodenal lymph node. However, we failed to identify the peritonitis seen repeatedly in recombinant RVFV-infected Ifnar1 2/2 mice. This suggests that, in Ifnar1 2/2 mice, cells of the macrophage lineage displayed an increase susceptibility to RVFV compared to wild-type mice. The high susceptibility of cells of the macrophage/dendritic lineage to viral infection in the absence of a functional type I IFN system has been previously observed. An increased infection of cells of the macrophage/dendritic lineage was observed in Ifnar1 2/2 mice infected with either the Sindbis virus [56] , or the mouse hepatitis virus [57] . Similarly, macrophages showed the greatest increase in susceptibility among the different splenocyte populations in West Nile virus-infected Ifnar1 2/2 mice compared to wild-type mice [58] . More generally, an increase in the replication of viruses in tissues and cells normally susceptible to virus infection has been previously observed in Ifnar1 2/2 mice. Coxsackievirus replicated dramatically in the liver of Ifnar1-deficient compared with wild-type mice [59] . Finally, previous investigation of poliovirus replication sites in infected Ifnar1 2/2 mice expressing the human poliovirus receptor showed that nontarget tissues became potentially permissive for virus infection when IFNa/b signaling was disrupted [60] . Therefore, the fact that Ifnar1 2/2 mice inoculated with NSsdeficient RVFV strains develop acute hepatitis and eventually die, as wild-type mice infected with a virulent RVFV strain, does not mean that the exact mechanisms of the cellular pathogenesis are the same in Ifnar1 2/2 and wild-type mice. Thus, although Ifnar1 2/2 mice have proven to be a tractable system in which to study the progression of RVFV infection in vivo, the immunocompromised nature of this mutant strain remains a limitation in translating these results directly to wild-type mice. Additional work needs to be done to develop similar whole-organ imaging and flow cytometry analysis in immune-competent mice. Further studies involving the use of fully virulent RVFV -i.e. carrying the NSs gene -which express a reporter gene, might allow us to give a comprehensive picture of the dynamics of natural infection in mammals. Our work provides the basis for the use of bioluminescent and fluorescent RVFV to study the effects of specific mutations in the viral genome and of host genetic factors on the tissue tropism and replication kinetics in living mice.

Evaluation of Internal Reference Genes for Quantitative Expression Analysis by Real-Time PCR in Ovine Whole Blood

The use of reference genes is commonly accepted as the most reliable approach to normalize qRT-PCR and to reduce possible errors in the quantification of gene expression. The most suitable reference genes in sheep have been identified for a restricted range of tissues, but no specific data on whole blood are available. The aim of this study was to identify a set of reference genes for normalizing qRT-PCR from ovine whole blood. We designed 11 PCR assays for commonly employed reference genes belonging to various functional classes and then determined their expression stability in whole blood samples from control and disease-stressed sheep. SDHA and YWHAZ were considered the most suitable internal controls as they were stably expressed regardless of disease status according to both geNorm and NormFinder software; furthermore, geNorm indicated SDHA/HPRT, YWHAZ/GAPDH and SDHA/YWHAZ as the best reference gene combinations in control, disease-stressed and combined sheep groups, respectively. Our study provides a validated panel of optimal control genes which may be useful for the identification of genes differentially expressed by qRT-PCR in a readily accessible tissue, with potential for discovering new physiological and disease markers and as a tool to improve production traits (e.g., by identifying expression Quantitative Trait Loci). An additional outcome of the study is a set of intron-spanning primer sequences suitable for gene expression experiments employing SYBR Green chemistry on other ovine tissues and cells.

To date, quantitative real-time PCR (qRT-PCR) is the most reliable and easy to perform technique to measure the expression level of a selected gene of interest (GOI) by quantifying mRNA transcripts. qRT-PCR is fast and the sensitivity of the method allows precise quantification of minimal differences in expression across a wide dynamic range even when working with limited amounts of starting material. However, several variables associated with the different steps of qRT-PCR experimental procedures can lead to considerable inter-sample variation and possibly to erroneous results: the different amount and quality of starting material; RNA integrity; efficiency in cDNA synthesis and PCR amplification; and differences between tissues or cells in overall transcriptional activity [1] . Among the strategies proposed to control for technical and sample variation in qRT-PCR experiments [2] , the use of reference genes is commonly accepted as the most reliable approach to normalize qRT-PCR and to reduce possible errors generated in the quantification of gene expression. In this normalization strategy, reference genes are used as internal controls and are submitted to the same experimental protocol of the GOI. The expression level measured for the target gene is then normalized according to the values of the internal controls. It is clear, therefore, that an ideal reference gene should be stably expressed within the samples to be compared irrespective of experimental conditions or external factors, otherwise the detection of small changes become unfeasible and unreliable. A number of studies have well assessed that genes classically thought to be stable for their ubiquitous expression and involvement in cell homeostasis (e.g., GAPDH, ACTB, 18S rRNA) are not always the best reference genes, as they show different behaviour across various cell types and tissues [3, 4] . Accordingly, a proper evaluation of several candidate genes should be performed before any gene expression study [2] . Studies aimed at identifying the most suitable reference genes in the ovine species have been performed in nervous tissues, spleen, mesenteric lymph node, ileum, lung and pulmonary artery [5] [6] [7] [8] . However, no specific information on whole blood is currently available. A reference gene for use in peripheral blood mononuclear cells was selected [9] , but this study was based on the analysis of the standard deviation of cycle threshold (C t ) and not on specifically designed algorithms. Nevertheless, blood is a readily accessible source of material for analysis and some attempts to identify gene expression markers by qRT-PCR in order to develop blood tests in sheep have been reported for prion diseases. For example, in 2001, Miele et al. discovered a novel erythroid-associated factor (ERAF) and demonstrated a dramatic decrease in expression of the specific transcript within rodent models of prion diseases, providing the first easily detectable molecular marker in a readily accessible tissue [10] . More recently, analysis of blood by qRT-PCR from sheep experimentally infected with scrapie revealed that the extent of differential expression of ERAF in peripheral ovine blood may be insufficient to provide a discriminatory diagnostic test [11] . However, the lack of a set of validated reference genes for sheep whole blood did not allow for the proper normalization of gene expression data and, in this study, glycophorin C (GYPC) was arbitrarily chosen as normalizer based on its higher expression in the human erythroid lineage. Moreover, the use of a single gene to normalise expression is no longer considered sufficient [12] [13] [14] [15] . Vandesompele et al. (2002) demonstrated that errors of up to 20-fold in expression data can be generated by the use of only a single reference gene [1] . The aim of the present study was to identify a set of reference genes to be used for normalizing qRT-PCR from ovine whole blood. We designed 11 PCR assays for commonly employed reference genes belonging to various functional classes and then determined their expression level in whole blood samples from control and disease-stressed sheep, both separately and combined, in order to select genes whose stability was unaffected under stress conditions. The geNorm and NormFinder applets [1, 16] were used for validating the reference genes; sample processing and experiments were carried out according to the Minimum Information for Publication of Quantitative Real-Time PCR Experiments (MIQE) guidelines [17] . Preliminary qRT-PCR experiments carried out to set up optimal reaction conditions showed that all candidate reference genes were expressed in ovine whole blood. qRT-PCR optimisation was performed using pooled cDNA samples in parallel with sheep genomic DNA. All primer pairs spanned two exons, generating melt-curve profiles specific to cDNA and genomic DNA amplification. While this strategy entailed much more effort in primer design and reaction optimisation, it assured specific amplification of mRNA transcripts by avoiding/recognizing interference of genomic DNA in quantification. Actually, in most cases DNAse treatment does not completely eliminate genomic DNA contamination, especially when RNA extraction is performed using reagents based on mono-phasic solution of phenol and guanidine isothiocyanate (personal observation). Gene-specific amplification was confirmed for all selected genes by a single-peak in melt-curve analysis and subsequent sequencing of amplicons. The determined reference gene sequences have been submitted to the GenBank database under the accession numbers JN811677-JN811687. The highest expression was obtained with ACTB, B2M and RPL19 with C t averages of 14.64, 15.82 and 15.95, respectively, whereas the lowest expressed gene was GYPC (mean C t , 27.50). For all analysed genes, the relative standard curve gave correlation coefficients greater than 0.985 and efficiencies between 90 and 110%. To select the optimal set of reference genes, expression values of the candidate genes were submitted to analysis by the geNorm and NormFinder applications. Table 1 reports the expression stability values (M) of the candidate reference genes in control, disease-stressed and combined sheep groups as calculated by the geNorm applet. The M values are used to rank genes on the basis of their stability: high M values indicate increased gene expression variability, whereas the most stable genes should exhibit M values <1.5 [1] . All studied genes reached acceptable stable expression with low M values, less than 1.5. Based on M value ranking, SDHA appeared to be the most stably expressed gene in control sheep with an average M value of 0.316, followed by YWHAZ and HPRT. In disease-stressed sheep, the overall stability of the candidate genes was lower and YWHAZ was the most stably expressed gene with an average M value of 0.624, followed by GAPDH and SDHA. When data from the two groups were combined, SDHA and YWHAZ To determine the optimal number of reference genes needed to calculate a normalization factor (NF), geNorm measures the pairwise variation between two sequential NFs with an increasing number of reference genes. A cut-off value of 0.15 is usually considered acceptable; it indicates that the control gene combination ensures satisfactory stability and that an additional gene need not be included. In the panel of candidate genes studied here, the use of two genes as references proved to be sufficient for accurate normalization in all sheep groups (Figures 1B, 2B and 3B). NormFinder ranks a set of candidate genes according to their expression stability measure (ρ) based on the similarity of their expression profiles. Lower values are assigned to the most stable genes. Table 2 reports the results of the NormFinder analyses. The ranking appears to be consistent to the one previously determined using geNorm. SDHA, YWHAZ and HPRT still occupy the highest positions in control animals, with stability values of 0.068, 0.125 and 0.132, respectively; while YWHAZ, GAPDH and SDHA shows the highest stability values in disease-stressed sheep (ρ values = 0.046, 0.064 and 0.188, respectively). When expression data from control and disease-stressed animals were combined, the resulting ranking confirmed SDHA and YWHAZ in the top positions with stability values of 0.093 and 0.096, respectively, followed by ACTB (ρ value = 0.099). TFRC and PGK1 are equally defined as the least reliable controls by both software and in all sheep groups. We examined the expression of 11 genes in ovine whole blood by using two commonly accepted softwares (geNorm and NormFinder). Both software algorithms are frequently used and freely available but have a different working rationale. NormFinder selects out of a set of potential reference genes one single best-performing reference gene that shows the least variation within the analysed group. GeNorm focuses on pairwise comparisons of reference gene expression in the experimental samples and so is less appropriate for identifying co-regulated genes [18] . To avoid possible bias, we therefore selected the candidate reference genes on the basis of differences in their physiological functions. To investigate the influence of the animal health status on the stability of the candidate reference genes, the analyses were performed in whole blood of control sheep and of sheep showing disease symptoms after clinical evaluation. Moreover, disease-stressed animals were sampled and analysed twice in order to monitor gene stability in disease-stressed sheep not only at different time points, but also under heat stress conditions, which, in association to disease status, really represent an extreme situation (see the Experimental Section for details on animal selection and sampling procedure). In all sheep groups, the results obtained with geNorm and Normfinder were consistent although not identical, as similarly reported elsewhere [19] [20] [21] [22] . SDHA and YWHAZ can be considered the most stably expressed genes in ovine whole blood ranking at the top positions in the control and disease-stressed sheep, both when they were analysed separately and when they were combined. SDHA and YWHAZ stability appears to be reliable as it was affected neither by disease status alone nor in association with heat stress. Moreover, geNorm indicated SDHA/YWHAZ as the best reference gene combination in the control and disease-stressed sheep joined datasets, a situation likely to fit most experimental contexts involving case and control animals; however, the SDHA/YWHAZ combination would be suitable for normalization of gene expression data also in studies carried out under physiological conditions, as these two genes demonstrated high stability in the control sheep group as well. Although data in sheep are still limited, SDHA appears to have good stability in this species as it was included in the reference genes required for reliable normalisation in several tissues (cerebrum, spleen, mesenteric lymph node and ileum). Similarly, YWHAZ was included in the optimal panel of reference genes to be used in the cerebellum, obex and ileum [5] . HPRT expression stability was evaluated only in the lung and pulmonary artery of brainstem death and control sheep, but it performed poorly as reference in both tissues on separate and combined analyses [7] . Actually, HPRT ranked as the third most stable gene of the control group in sheep whole blood, but its stability strongly decreased under disease conditions, emphasizing that proper validation of reference genes in a cell type or tissue of interest and under different experimental settings is mandatory before reporting qRT-PCR results. B2M showed stable expression in one study on human leukocytes from 13 healthy donors [1] . B2M also had stable expression in a large study in which 526 human whole blood samples represented healthy individuals and six disease groups [23] . In sheep, however, B2M is outperformed by other genes and demonstrates suboptimal suitability as reference gene in whole blood. In humans, GYPC expression is considerably higher in erythroid lineage cells than in non-erythroid cells [24] . GYPC was therefore used by Brown et al. (2007) to normalize qRT-PCR analyses of erythroid markers in sheep whole blood [11] . In our study, however, both geNorm and NormFinder classified GYPC in the bottom half of the stability ranking under both control and disease conditions, showing that ubiquitously expressed genes would provide a more relevant comparison for measuring erythroid gene expression. In control sheep, GAPDH resulted in being the gene with the highest degree of individual variation in expression level. This finding was not surprising, as GAPDH can be regulated under a large number of physiological states and is generally not considered a good reference gene [25, 26] . Nevertheless, GAPDH ranked as the second most stable gene in the disease-stressed and the combined sheep groups. Taylor et al. (2008) found that the levels of GAPDH transcription were the most stable of the genes tested in ovine peripheral blood mononuclear cells during infection with Mycobacterium avium subsp. paratuberculosis [9] . This finding is consistent with our results indicating stability of GAPDH in whole blood under disease condition. However, it must be taken into account that Taylor's results could also be attributable to a characteristic of the specific cell type or to the fact that the study based its observations on analysis of the standard deviation of C t values and not on specifically designed algorithms. A distinctive point of our study is the major effort put into primer design with the aim to validate only oligos spanning at least one intron. This aspect has been neglected in the previous works on reference gene validation in sheep, probably because of the lack of ovine genomic DNA sequences available in the public databases. Indeed, we were able to retrieve intron-spanning primers from previous publications for only three genes (PGK1, SDHA and G6PD) among those included in our study. Nevertheless, this approach is highly recommended in combination with DNAse I treatment to avoid/recognize co-amplification of contaminating genomic DNA [1] , since spurious PCR signals could affect the selection of reliable references by mimicking individual variation with lower stability scores. Also, when searching RTPrimerDB, a reference database for qRT-PCR primers [27] [28] [29] [30] , we noted that among 8329 real-time PCR primer sets for 5758 genes of 26 organisms available at the time of writing, only 16 SYBR Green assays were deposited under Ovis aries. Importantly, therefore, an additional outcome of our study is a set of validated primer sequences suitable for gene expression experiments based on SYBR Green chemistry to be carried out with other ovine tissues and cells. Fresh whole blood samples were collected into EDTA tubes from 28 Biellese sheep belonging to three different farms. The animals included in the study were unrelated. Sheep were submitted to clinical evaluation by a veterinarian and categorized as control animals (n = 18), not showing any clinical sign, and disease-stressed animals (n = 10). Specifically, they had chronic diarrhoea (n = 5), lameness (n = 2), abscesses (n = 2) and respiratory syndrome (n = 1). The blood samples from disease-stressed sheep were collected twice: the first sampling was carried out in August, when the environmental temperature was of 35 °C, and the second in September with an environmental temperature of 26 °C. Every sampling was preceded by clinical evaluation of sheep to confirm disease status. The blood samples were immediately transferred to the lab and submitted to nucleic acid isolation. Total RNA was extracted using the QIAamp RNA Blood Mini Kit (Qiagen) according to the manufacturer's instructions. Contaminating genomic DNA was removed by on-column treatment of each sample with DNase I (Qiagen). Purity, concentration and integrity of total RNA were assessed using two independent techniques. RNA purity and concentration were evaluated by absorbance readings using a NanoDrop ND-1000 spectrophotometer (Thermo Fisher Scientific). RNA quality was determined with an RNA 6000 nano LabChip Kit in the Agilent Bioanalyzer 2100 system. Quality was evaluated using the RNA Integrity Number (RIN) [31] . The mean total RNA concentration was 96 ng/µL while A260/A280 and A260/230 ratios ranged from 1.99 to 2.04 and 2.02 to 2.16, respectively. Therefore all samples were pure, free from protein and organic pollutants derived from RNA extraction. The RIN obtained for all samples ranged from 7.2 to 8.2 with a mean value of 7.6. Total RNA (500 ng) was reverse transcribed using a High Capacity cDNA Reverse Transcription Kit (Applied Biosystems) according to the manufacturer's protocol in a final volume of 20 µL. The cDNA was subsequently stored at −20 °C. Pooled cDNAs were then used in preliminary experiments to evaluate primer performance and specificity and for PCR protocol optimization. Subsequently, the expression profile of the selected genes was analysed in each cDNA sample separately. We selected 11 genes belonging to various functional classes and frequently used as references in qRT-PCR gene expression experiments: β-actin (ACTB); tyrosine 3-monooxygenase/ tryptophan 5-monooxygenase activation protein, zeta polypeptide (YWHAZ); hypoxanthine phosphoribosyl-transferase I (HPRT); transferrin receptor (TFRC); succinate dehydrogenase complex, subunit A (SDHA); β-2-microglobulin (β2M); phosphoglycerate kinase I (PGK1); glyceraldehyde-3phosphate dehydrogenase (GAPDH); glucose-6-phosphate dehydrogenase (G6PD); ribosomal protein L19 (RPL19); and glycophorin C (GYPC). Specifically, GYPC was included into the panel because of its expression in the erythroid lineage and because it was used in a previous study as normalizer to quantify the expression of erythroid genes in sheep blood [11] . Primers for PGK1, SDHA and G6PD were based on previous publications [5, 7] . The other primers were designed using Primer3 software [32] by aligning ovine sequences available in GenBank with bovine and human homologous genes. Primers were selected to produce amplicons spanning two exons and their specificity was tested using ovine pooled cDNA and genomic DNA in preliminary PCR assays. The PCR products were subsequently run on 2% agarose gel to check for size specificity and, eventually, sequenced. Table 3 summarises primers information including sequences, product size, putative exon position, estimated size of the amplicon, efficiency of RT-PCR (E) and correlation coefficients (R 2 ). All PCR reactions were performed in a 25-µL final volume containing 2× Brilliant II SYBR Green Master Mix (Stratagene), 300-900 nM of each specific primer and 1 µL of cDNA. PCR amplification was run on a Mx 3005P QPCR System (Stratagene) using 96-well optical plates under the following conditions: 10 min at 95 °C for polymerase activation, and 40 cycles of 3-segment amplification with 30 s at 95 °C (for denaturation), 30 s at 56-60 °C, and 40 s at 72 °C for elongation. Primer concentration and annealing temperatures were optimised to individual genes; specifications are available from the Authors upon request. A dissociation step was added after elongation to ensure that the desired amplicon was detected. The dissociation step eliminates non-specific fluorescence signal and ensures accurate quantification of the desired product. Finally, a melting curve was produced to confirm single gene-specific peaks and to detect primer/dimer formation by heating samples from 60 to 95 °C. PCR efficiencies were calculated using a relative standard curve derived from a pooled cDNA mixture (a 10-fold dilution series with five measuring points). All experiments were replicated twice for each gene with triplicate sample runs within each replication and a no-template control was included using water instead of cDNA. qRT-PCR data were analysed for reference genes expression stability using two different statistical algorithms: geNorm version 3.5 [1] and NormFinder version 0.953 [16] according to the developers' recommendations. Raw quantification cycle (C t ) values were converted to relative quantities using the comparative C t method as input data for the two applets. Preliminary analyses performed separately on qRT-PCR data from disease-stressed animal sampled at different time points and under different environmental temperatures (August and September) retrieved consistent results. Therefore C t values were averaged after inter-run calibration according to Hellemans et al. [33] and submitted to subsequent data analysis. The combined analysis was performed processing samples from ten randomly chosen control sheep together with the disease-stressed sheep and then joining expression results in a combined dataset. A number of studies have been carried out to identify reliable reference genes in specific tissues in various species [34] [35] [36] [37] [38] . In sheep, analyses of expression stability of candidate reference genes are limited to a restricted range of tissues [5] [6] [7] [8] and data on ovine whole blood were still lacking. However, peripheral whole blood is attractive because of its accessibility and usefulness in monitoring several physiological and pathological conditions. As regards disease status, blood certainly represents the best tissue for in vivo test development since collection is non-invasive and easy to perform. For example, the identification of differentially expressed genes acting as indirect in vivo markers in blood would represent a major breakthrough for the diagnostics of non-conventional agents (like prions) which currently cannot be detected by standard methods as the diagnosis still rely on post mortem investigations. This study provides a panel of optimal control genes for use in qRT-PCR studies in sheep whole blood. The two softwares tested, based on different algorithms and analytical procedures, produced highly comparable results. SDHA and YWHAZ represent good reference genes for gene expression studies in sheep peripheral whole blood, unaffected by disease status and heat stress conditions, and the geometric mean of these two stable genes is an accurate normalization factor [1] . Our results may be useful for the identification of genes differentially expressed in a readily accessible tissue, with the potential of discovering new physiological and disease markers and as a tool to improve production traits (e.g., by identifying expression Quantitative Trait Loci (eQTLs) [39, 40] .

Normal variation in thermal radiated temperature in cattle: implications for foot-and-mouth disease detection

BACKGROUND: Thermal imagers have been used in a number of disciplines to record animal surface temperatures and as a result detect temperature distributions and abnormalities requiring a particular course of action. Some work, with animals infected with foot-and-mouth disease virus, has suggested that the technique might be used to identify animals in the early stages of disease. In this study, images of 19 healthy cattle have been taken over an extended period to determine hoof and especially coronary band temperatures (a common site for the development of FMD lesions) and eye temperatures (as a surrogate for core body temperature) and to examine how these vary with time and ambient conditions. RESULTS: The results showed that under UK conditions an animal's hoof temperature varied from 10°C to 36°C and was primarily influenced by the ambient temperature and the animal's activity immediately prior to measurement. Eye temperatures were not affected by ambient temperature and are a useful indicator of core body temperature. CONCLUSIONS: Given the variation in temperature of the hooves of normal animals under various environmental conditions the use of a single threshold hoof temperature will be at best a modest predictive indicator of early FMD, even if ambient temperature is factored into the evaluation.

Foot-and-mouth disease (FMD) is a highly infectious viral disease of cloven-hoofed animals, both domestic and wild. The disease is caused by a small RNA virus, which is 28 nm in diameter and exists as seven serotypes. The disease is characterised by fever, and blisters in the mouth, on the feet and on the teats and these rupture and are associated with slobbering and lameness. Adult animals may suffer weight loss and milk production can decline significantly. Though most animals eventually recover from FMD, the disease can lead to myocarditis and death, especially in newborn animals [1] . FMD is found regularly in parts of South America, Africa, the Middle East and other parts of Asia and periodically spreads to affect normally disease free countries. It is a significant impediment to trade in livestock and their products as countries with the disease face restrictions for exporting to disease free regions. Moreover, the disease is difficult and costly to control and eradicate. The Royal Society [2] estimated that during the 2001 epidemic in the UK, in which some six million animals were culled, the losses to agriculture and the food chain were £3.1 billion and some £2.5 billion was paid by the UK Government in compensation for slaughtered animals and clean-up costs. Losses were also experienced in tourism and business directly affected by tourism; it has been estimated these were between £2.7 and £3.2 billion [3] . Two other epidemics highlight the global impact of the disease; the first a major epidemic in Argentina in 2001 and the second in Japan during 2010; in the first two thousand five hundred and nineteen herds were infected [4] and in the second two hundred and fifty (Office international des épizooties-World Organisation for Animal Health, 2010. Follow-up report Early identification of animals infected with FMD virus is vital if disease outbreaks are to be rapidly diagnosed and controlled. Thorough screening to identify signs of FMD is time consuming and labour intensive since it requires the capture and restraint of suspect animals for clinical examination. This can be particularly difficult in some situations, for example where animals are at pasture, are difficult to handle or are present in very large numbers. Animals with FMD often develop a fever with temperatures in excess of 40°C and vesicular lesions around the coronary band, in and around the mouth and on the mammary gland. The vesicular lesions are associated with local inflammation giving rise to an increase in skin temperature which can be detected by palpation [1] . On their own, these temperature changes are not pathognomonic for FMD but can be used to select animals that warrant closer examination to detect more definitive signs and/or enable sampling for confirmatory testing. Infrared thermography (IRT) can be used to measure the heat emitted from a surface and to display and store an image and associated data. The technique has been used by the medical profession over recent years across a range of human conditions, to identify local inflammations or pyrexia [5] and in the detection of fever associated with SARS and avian influenza [6] . IRT has also been used by those involved with animal disease [7] [8] [9] . Workers at the Pirbright Laboratory of the Institute for Animal Health (IAH-Pirbright) and at the Plum Island Animal Disease Center (PIADC), USA have reported that IRT can be used to measure the temperatures of animals that need to be checked for possible onset of FMD [10] [11] [12] . These workers studied groups of animals with experimentally-induced FMD and measured temperatures (primarily around the coronary band) as disease progressed. It was found that increases in temperature associated with FMD could be detected, sometimes prior to the development of visible lesions. Unpublished work by the current authors involving five cattle, five sheep and five pigs infected with the Asia 1 strain of FMDV discovered that it was easy to measure the feet temperatures of the animals and established that there was potential for using the technique in the field. Cattle feet temperatures ranged from 18.7°C to 31.7°C, with the highest value being recorded the day before foot lesions were visible, but at the same time as the first lesion on the tongue. Prior to the first appearance of lesions temperatures were below 27°C. To optimise interpretation of temperature measurements and to demonstrate the reliability of the technique to differentiate between infected and healthy livestock requires further IRT data from uninfected animals, kept at different ambient temperatures and under different husbandry conditions. This shortcoming is addressed here by IRT measurements and analysis from healthy cattle. The experimental period was divided into two phases. The first phase was designed to make observations under different IRT/animal configurations and environmental conditions, the second to examine the changes in an animal's hoof temperature over a daily cycle of activity. In the first phase, five separate sets of temperature data were taken over a period of five months using a TIR1 imager manufactured by Fluke (temperature range -20°C to 100°C, accuracy +/-2°C, operated at a distance of 1 to 2 m, emissivity 0.95). Two groups of nine and ten cattle initially aged 12 and 3 months old respectively, were housed in small groups in pens in an open barn at the Institute for Animal Health farm at Compton, Newbury; one half of each pen had a concrete floor and the other a slightly raised straw filled area. Each animal in turn was restrained either by hand or in an animal crush and four IRT measurements were taken of each of the animal's feet from different aspects (front, back, lateral and medial) and a measurement was also taken of the left eye (see Figure 1 for an example). These two sites were selected because, as mentioned above, researchers working on FMD had detected an increase in temperature around the coronary band and it is hypothesised that temperatures around the eye provide a non-invasive indicator of an animal's core temperature. Other sites commonly affected by FMD lesions such as the mouth and udder are less accessible and/or only applicable to lactating animals. To investigate the link between eye temperature and body temperature each animal's rectal temperature was taken at the same time as the thermal images using a digital thermometer. The IRT images were taken twice within ten minutes from each animal to evaluate repeatability of the measurements and correct for minor variations in the angle of the imager to the animal. Ambient temperatures were measured with a Fisher Scientific model FB70357 digital thermometer. Care was taken throughout the experiment when handling the cattle, as it was appreciated that even the simple act of gathering animals can cause an increase in stress which in turn may result in an increase in the animal's temperature. As it was not practical to measure changes in an animal's hoof temperature over an extended period of activity using an IRT imager, a second temperature measuring device was used for this purpose (IButton data loggers, type DS1921G, temperature range -40°C to 70°C, accuracy +/-1°C, data recording rate every second or every two seconds, manufactured by Embedded Data Systems). The IButtons were strapped to the animals' hooves as shown in Figure 1 . To compare the results of IRT and IButtons, hoof temperatures were measured for two cattle. TIR1 images were taken either immediately prior to IButton attachment, simultaneously with attachment, but for another foot, or immediately after the IButton was removed. Identical readings from the two instruments were not expected as both devices measure temperature in different ways (TIR1-radiative and IButton-thermal contact). However, it was anticipated that similar trends could be detected using both sensors. The final phase of the work was to investigate changes of hoof temperature as a function of activity. These were established using IButtons and a video security camera (Solidex Day Night DomeCam Varifocal Lens combined with a Solidex 4 channel DVR) placed above the pen holding two of the cattle (chosen for ease of visual recognition). IButtons were attached to the two hind feet of the cattle and data recorded at a frequency of once or twice per minute for a period of approximately twenty hours. Air temperatures were recorded using an IButton suspended in free air close to the animal pen. The experiment was done twice. TIR1 images were processed using Smart View Software (V2.1.0.10), supplied by Fluke. For each image, an area of approximately 2 cm 2 above and below the coronary band was selected and the maximum temperature within this area (see Figure 1 ) was recorded and transcribed to an Excel spreadsheet for subsequent statistical analysis. Additionally, an area at least 10 cm above the hoof was selected to determine whether a ratio between hoof and more proximal leg surface temperature could help compensate for hoof temperature changes caused by ambient temperature changes. An area of approximately 2 cm 2 around the eye was selected for analysis and the hottest temperature within this area, including the eye itself, was recorded (see Figure 1 ). For comparing IButtons and IRT, the area covered by the attached IButton was selected and the average temperature within this area recorded and further analysed in an Excel spreadsheet. IButton data was analysed with TempIT software supplied by Signatrol (Version 4.1.8) and the data transferred to the master Excel spreadsheet. To determine the animal's movements the security camera images were replayed and activity allocated into one of four categories (lying down, standing on deep straw, standing on concrete and outside of the holding pen). The date and time for each change in activity category was recorded for comparison with the IButton data. Two separate analyses of the data were carried out to assess: (i) the repeatability of thermal image measurements taken sequentially within a ten minute interval; and (ii) the potential for defining a threshold temperature above which cattle would be considered abnormal based on IRT. The repeatability of thermography was assessed by computing the difference in temperature as measured by corresponding images (i.e. for the same hoof with the same aspect on the same day) for each animal and determining whether the median differed significantly (P < 0.05) from zero using a Wilcoxon signed rank test. The potential for defining a threshold temperature to identify unhealthy cattle based on IRT was examined using a Bayesian hierarchical model, which incorporates between-animal variation and facilitates predictions outside the data which allow for parameter uncertainty. In this approach, the observed hoof temperature (T jk ) for the jth observation on animal k was described by, jk is the expected hoof temperature, σ 2 e is the error variance, the b (k) i s are parameters and X ijk is the value of the ith factor (e.g. hoof, aspect or ambient temperature) for the jth observation on animal k. Betweenanimal variation was modelled by assuming that the parameters for each animal are drawn from higherorder distributions, such that, Non-informative priors were used for the higher-order parameters: diffuse Normal distributions for the μ b is and diffuse gamma distributions for the σ b is. Parameters in the model were estimated using Markov chain-Monte Carlo methods implemented in WinBUGS [13] . Two chains of 50,000 iterations were run for each model, with the first 10,000 iterations discarded to allow for burn-in of the chain. Each chain was then thinned by sampling every tenth iteration to reduce autocorrelation amongst the samples. The fits of different models were compared using the deviance information criterion (DIC) [14] . Posterior predictions for the expected hoof temperature as a function of ambient temperature were generated by sampling from the joint posterior density for the higher-order parameters. A range of percentiles of the resulting distribution were used to define thresholds for identifying abnormal animals and the specificity of a classification scheme based on these thresholds (essentially the proportion of animals below the threshold) was assessed. In the first phase of the experiment, between July and November 2009, around two thousand three hundred thermal images of cattle hooves were taken. During these experiments, ambient temperatures ranged from 10°C to 24.8°C and general weather conditions from a warm summer's day through to cold and damp winter conditions. Hoof temperatures measured by IRT ranged from approximately 10°C to 36°C (Figure 2a ) and depended markedly on ambient temperature (Figures 2d &3) . Furthermore, the variability in hoof temperatures was greatest at lower ambient temperatures (Figure 2d ). The median range in hoof temperatures for individual animals on a given day was approximately 6°C, but in some cases it was > 12°C (Figure 3 ). This range often reflected one hoof or side being markedly warmer than the other (for example, animals 362, 762, 766 and 769), but sometimes there was no clear explanation for the difference (for example, animals 763 and 777). Differences in hoof temperature between corresponding images recorded on the same day were typically small ( Figure 2b ) and did not differ significantly (P > 0.05) from zero for 13 (out of 19) animals. For six animals (animals 379, 762, 766, 768, 769 and 774), the median difference in repeated observations was significantly (P < 0.05) different from zero, though the median difference in each case was only a fraction of a degree (range: -0.3°C to 0.2°C). Eye temperature measured by IRT provided a reasonable proxy measure for body temperature, with eye temperatures approximately 2°C lower than rectal temperature ( Figure 2c) and not significantly affected by ambient temperature (P > 0.05). An adequate model to describe the hoof temperature data included ambient temperature (°C), hoof (coded as: front-left, front-right, hind-left and hind-right) and camera aspect (coded as: front, lateral, medial and rear) ( Table 1) ; removing any of these terms from the model significantly worsened model fit (full model: DIC = 17270.6; removing ambient temperature: DIC = 18846.8; removing hoof: DIC = 17309.8; removing aspect: DIC = 17297.2). Adding extra terms to the model improved the model fit, often markedly so; for example, a quadratic term for ambient temperature (DIC = 14730.3) or eye temperature as a normalising factor (DIC = 16648.1). However, this was at the expense of the resulting model being poor as a predictive tool, because the variance for the higher-order parameters needed to be so large to incorporate the observed differences amongst animals. Accordingly, the adequate model was used in subsequent analyses. The analysis indicated that there was variation in temperature amongst hooves on the same animal, but these differences were not systematic between animals, as evidenced by means for the hoof parameters which do not differ significantly from zero, but which have a high standard deviation (Table 1) . Camera aspect did influence hoof temperature measurement, with images taken from the lateral, medial or rear aspect being around 1°C lower than those taken from a front aspect (Table 1) . However, ambient temperature had the greatest impact on hoof temperature (Figures 2d &3; Table 1 ). By sampling from the joint posterior density for the higher-order model parameters (and integrating out the effects of hoof and camera aspect) it was possible to generate predictions for hoof temperature as a function of ambient temperature. The 75th, 90th and 95th percentiles for these predictions were then used to define thresholds by which to identify healthy cattle, with a further refinement that the maximum threshold temperature was set equal to the mean rectal temperature for the animals (38.3°C) (Figure 4a ; Table 2 ). The specificity of a classification scheme based on these thresholds was investigated. For a threshold based on the 75th percentile, the predicted specificity was low, especially at ambient temperatures below 20°C (< 80% specificity, Figure 4b ). The specificity was improved by setting a threshold based on the 90th or 95th percentile with specificity > 90% predicted above temperatures of 15°C and 10°C respectively (Figure 4c, d) . A simple comparison between the TIR1 and three IButtons, on a shaded uniform temperature carpet tiled floor revealed that both instruments recorded similar temperatures with the TIR1 being warmer than the IButton by 0.1 to 1.4°C (IButton no./TIR1/IButton: 1/24.3/23.0; 2/24.3/ 22.9; 2/23.6/22.9; 3/24.3/23.5; 23.6/23.5°C). It was also established that the IButtons, given a sudden temperature change of 15°C took fifteen minutes to reach equilibrium. Table 3 presents the results from the comparison between the IButton and TIR1 for two animals on two separate days. The data show that temperature measured by the IButton approximately fifteen minutes after attachment and just before removal relates well to the average temperatures measured by the TIR1. The IButton temperatures were consistently warmer than the average temperature measured by TIR1 (average temperature differences 5.3, 5.8, 4.4 and 1.9°C). This trend was observed for all data collected during the comparison of the IButton and TIR1. The IButton as well as the TIR1 record sudden temperature changes equally well as seen on one occasion where the average temperatures for all feet for one animal measured by the TIR1 were 10.2/11.2/12.3 and 9.6°C before IButton attachment; whereas after the removal of the IButton average temperatures for all legs measured by the TIR1 were only very slightly raised (0.1-1.3°C) apart for one foot where the temperature was raised by 14.7°C. The IButton recorded 14°C and 15°C for three out of the four legs at fifteen minutes after attachment and before removal, but recorded temperatures of 18°C after fifteen minutes and 29.5°C before removal for the leg with the raised temperature (data not shown). The extended measurement period using both the IButton and security surveillance camera showed that the animals' hoof temperatures varied by as much as 20°C depending upon a combination of activity and ambient air temperature. A typical analysis is given at Figure 5 where it can be seen that when the animal was standing on the concrete temperatures were much lower than when it was lying down in the straw with its feet tucked under its body. This effect was consistent in each of the animals whose temperatures were measured. Maximum temperatures of 38°C were recorded and this was very close to the animal's rectal temperature. To detect inflammatory conditions such as FMD affecting cattle feet, IRT needs to be able to identify abnormal surface temperature elevations. This raises the challenge of being able to distinguish such elevations from the spectrum of variability found in uninfected animals. Similar challenges affect the use of the technique in screening human subjects, for instance for pyrexia at airports [6] . Two approaches can be envisaged for FMD. First, IRT could prove very useful if a threshold temperature were to be established above which a foot temperature triggers a suspicion of an inflammatory condition. This approach has been suggested by Rainwater-Lovett [12] . However, the current study shows that this technique may be too simplistic in its approach as an animal's hoof temperature is significantly affected by ambient temperature and posture/activity. Thermal image data reported by Bashiruddin [11] from FMDV-infected cattle were compared with the thresholds shown in Table 2 to determine if these thresholds could provide a basis for early stages of FMD infection to be detected. At the ambient isolation facility temperature of~16°C, Table 2 suggests that hoof temperatures of 30.5°C (75 th percentile), 34.9°C (90 th percentile) and 37.6°C (95 th percentile) would indicate an elevated temperature indicative of infection. However, of the five animals which became infected, only one showed a temperature above 30°C (two hooves) and this was when vesicular lesions were visible. Although definitive conclusions will require study of greater numbers of infected animals, these results suggest that the threshold temperatures determined in the present study will result in a low sensitivity, unless specificity is reduced. An alternative approach is for the operator to use IRT to identify hot-spots. These are identified as either part or all of a hoof that is hotter than the surrounding skin or hotter than other feet. In this approach, it is relative rather than absolute temperatures that matter. Previous studies [11, 12] have demonstrated that areas of raised temperature on an animal's hoof can be detected. To investigate this approach, a "blind test" was conducted using forty four thermal images from six cattle either before infection or in the early stages of FMD [11] . One of the authors was invited to categorise the images as "not a concern", "unlikely to be infected", "possibly infected", "suspicious" or "highly suspicious". Once an animal displayed clinical signs evident upon close physical examination, it was considered infected and it was excluded from further analysis the day afterwards, since temperatures of the feet often decline within a day or two of the formation of vesicles even if ruptured lesions remain evident. The results from this pilot revealed a 70% sensitivity (7 out of 10 images) and 79% specificity (scoring possibly infected and above as positive) (27out of 34 images) or 30% sensitivity (3 out of 10 images) and 94% specificity (scoring suspicious and above as positive) (32 out of 34 images). Whilst these results are encouraging, further work using images collected from a larger number of infected animals is needed before a conclusion can be reached concerning the merits of this approach. This study has been completed under ideal field conditions. The situation in the field is likely to be less favourable. For example the animal's feet may be wet, covered in grass, muddy or covered in faeces. These variables need to be studied in more detail before IRT can be used with confidence to detect FMD in the field. Other parts of the body affected by inflammation in FMD, such as the mouth are not readily visualised by an infrared camera, whilst changes in the udder are limited in application to female dairy breeds. The use of IRT eye measurements seems a promising method to measure body temperature and therefore merits further evaluation in animals affected with FMD and other pyrexic conditions. If IRT technology is to be useful in the field it has to be both technically capable of distinguishing between infected and non infected animals and be a cost effective diagnostic tool. In the field two scenarios are likely; the first where animals are housed or can be easily corralled and are readily accessible at close range and the second where they are at pasture and less easy to gather. In the first instance the current cost of an IRT camera will be in the range £2-10 k but in the second, where the equipment is required to operate at longer ranges, it is likely that a more powerful telephoto lens would be required. The cost of this significantly increases the price of the equipment possibly up to £20 k. The study has identified that an animal's hoof temperature is influenced by its activity prior to the point at which thermal screening is performed. Consequently, a period of acclimatisation is required prior to an image being taken. This is particularly the case if the animal has been lying down with its feet tucked under its body. The work has shown that IRT images of an animal's eye temperature may be a useful proxy for core temperature and could be used to detect pyrexia as an indicator for selecting animals for closer examination. This conclusion supports the observation by Dunbar [10] who compared high quality thermograms of the eye (n = 16) to body temperature and found them not to be different (p = 0.19). However, further work is required with animals infected with FMDV to confirm this.

Cochrane Systematic Reviews of Chinese Herbal Medicines: An Overview

OBJECTIVES: Our study had two objectives: a) to systematically identify all existing systematic reviews of Chinese herbal medicines (CHM) published in Cochrane Library; b) to assess the methodological quality of included reviews. METHODOLOGY/PRINCIPAL FINDINGS: We performed a systematic search of the Cochrane Database of Systematic Reviews (CDSR, Issue 5, 2010) to identify all reviews of CHM. A total of fifty-eight reviews were eligible for our study. Twenty-one of the included reviews had at least one Traditional Chinese Medicine (TCM) practitioner as its co-author. 7 reviews didn't include any primary study, the remaining reviews (n = 51) included a median of 9 studies and 936 participants. 50% of reviews were last assessed as up-to-date prior to 2008. The questions addressed by 39 reviews were broad in scope, in which 9 reviews combined studies with different herbal medicines. For OQAQ, the mean of overall quality score (item 10) was 5.05 (95% CI; 4.58-5.52). All reviews assessed the methodological quality of primary studies, 16% of included primary studies used adequate sequence generation and 7% used adequate allocation concealment. Of the 51 nonempty reviews, 23 reviews were reported as being inconclusive, while 27 concluded that there might be benefit of CHM, which was limited by the poor quality or inadequate quantity of included studies. 58 reviews reported searching a median of seven electronic databases, while 10 reviews did not search any Chinese database. CONCLUSIONS: Now CDSR has included large numbers of CHM reviews, our study identified some areas which could be improved, such as almost half of included reviews did not have the participation of TCM practitioners and were not up-to-date according to Cochrane criteria, some reviews pooled the results of different herbal medicines and ignored the searching of Chinese databases.

Traditional Chinese Medicine (TCM) is an essential part of the healthcare system in several Asian countries, and is considered a complementary or alternative medical system in most Western countries [1] . Chinese herbal medicines (CHM) are an essential part of TCM [2] . The 2002 National Health Interview Survey showed that 18.6% of adults used CHM in the United States, while it was 12.1% in 1997 [3] . With the increased use of CHM, questions arise from clinicians, patients, and policymakers as to the effectiveness of these interventions [4] . In an era of evidence-based healthcare, systematic reviews of randomized controlled trials (RCTs) are becoming increasingly important as a source of evidence for decision-making. As the number of systematic reviews of CHM increase, the quality of which has been highlighted and called into question. Some studies have assessed the quality of CHM reviews published in Chinese journals, in general, they have been criticized for lacking a comprehensive search for clinical trials, ignoring the characteristics of TCM, using inappropriate criteria to assess the methodological quality of included studies, and addressing too broadly defined questions [5] [6] [7] . All [8] these aspects could have contributed to a poor quality review. The Cochrane Collaboration is an international organization that aims to prepare and maintain rigorous systematic reviews in order to help people make well-informed decisions about health care [8] . Compared with reviews published in paper-based journals, Cochrane reviews are noted to have greater methodological quality [9] . Ever since 1999 when the first Cochrane review of CHM was published, a sharp increase has been observed in the number of similar reviews. However, no previous studies have systematically assessed the methodological quality of Cochrane reviews of CHM. Therefore, we did this overview of systematic reviews. Our study had two objectives: a) to systematically identify all existing Cochrane reviews of CHM; b) to assess the methodological quality of included reviews. Data for this study was acquired through previously published work, no patient or hospital data was accessed. Therefore, written consent and institutional ethical review was not required for this research. The PRISMA checklist and flow diagram are available as supporting information; see PRISMA Checklist S1 and PRISMA Flow Diagram S1. In order to identify reviews focusing on CHM, we searched the titles and abstracts of all reviews contained within the Cochrane Database of Systematic Reviews (CDSR) (Issue 5, 2010) using the following terms: Chinese or herb* or traditional or plant or medic*. We included all Cochrane reviews of CHM. Protocols and reviews which have been withdrawn from publication were excluded. We defined CHM as preparations derived from plants or parts of plants (e.g. leaves, stems, buds, flowers, roots or tubers) that grow in China and have been widely used for medical purpose. CHM include single herbs (or extracts from single herbs) and compound formulas of several herbs in all forms of preparation formulation (e.g. oral liquid, tablet, capsule, pill, powder, plaster or injection liquid). It should be noted that our definition of CHM does not include plant-derived chemicals or synthetic chemicals which contain constituents of plants. For example, although Huperzia serrata has its origin in China, according to our definition, Huperzine A does not belong to CHM because it is a kind of alkaloid extracted from Huperzia serrata. In addition, we only included reviews discussing herbs which originated from China, reviews on herbs such as Passiflora and Echinacea, both of American origin, were invariably excluded. Oxman-Guyatt Overview Quality Assessment Questionnaire (OQAQ) [10] The OQAQ instrument was selected as the quality appraisal tool, which was designed to evaluate whether the authors of a systematic review conducted a comprehensive search, minimized bias in the selection of primary studies, evaluated the primary literature, and pooled the results appropriately. It consists of 10 questions, the first 9 questions are designed to assess different aspects of methodological quality and have set answers of ''yes'', ''partially/can't tell'', or ''no'', question 10 is an assessment of the overall scientific quality of the systematic review on a scale of 1 to 7, it is answered based on how well the review scored on the first 9 questions. We established a database (using Microsoft Excel 2007) to extract data. The database had two components: 1) general characteristics, including country of first author and number of authors, whether the review had the participation of TCM practitioners, number of trials and participants included, disease, the year of review last assessed as up-to-date, conclusions drawn by the reviewers (by assessing the reviewers' abstract conclusions statements), interventions in experimental groups, number of herbs included, and whether the results of different herbal medicines were pooled; 2) methodological quality of included reviews, including OQAQ scale, the approach to assessment of methodological quality of primary studies, the number of trials with adequate sequence generation and allocation concealment, and type and number of English and Chinese databases searched. Two reviewers (Jing Hu and Wei Zhao) independently extracted the information of each review, disagreements between the two reviewers were resolved by discussion. The questions addressed by a review may be broad or narrow in scope, each review was assigned into one of the following two categories: 1) narrowly focused reviews, intervention in each review was single herb or herbal preparation, as an example of ''Chinese herbal medicine suxiao jiuxin wan for angina pectoris''; 2) broadly focused reviews, including reviews concerned multiple Chinese herbs or a family of herbal medicines sharing similar efficacy, such as ''Chinese herbal medicine for premenstrual syndrome'' and ''Chinese herbal medicine Huangqi type formulations for nephrotic syndrome''. For broadly focused reviews, we listed the number of herbs included and assessed whether the results of different herbal medicines were pooled. A review was believed to have the participation of TCM practitioners if at least one author works in TCM department, university or hospital, or it stated that it had got suggestion from TCM practitioners. When we assessed the type and number of English and Chinese databases searched, we only listed the databases which at least 4 reviews searched. In addition, the Cochrane Specialized Register and databases/websites for ongoing trials were also searched in some reviews, we did not list them in our study. 278 potentially relevant reviews were obtained, after selection (according to inclusion and exclusion criteria), a total of 58 Cochrane reviews were eligible for our study, a full list of reviews is included in Table S1 . Of the 58 reviews, one review [49] included herbs originated in China, India and Japan; interventions in another review [55] concerned both herbal and chemical medicines. In these two cases, we extracted and analyzed the information relating to CHM. The number of authors in the 58 reviews ranged from 1 to 10, the first authors were most often from China (46 [79%]), followed by UK (n = 8), and Netherlands, Canada, USA and Australia each have one first-authored review. Twenty-one (36%) of the included reviews had at least one TCM practitioner as its co-author. 7 (12%) reviews didn't include any primary study, of the remaining reviews (n = 51), a total of 671 studies and 75,609 participants were included, the median number of studies and participants included were 9 (Quartile: 3, 15) and 936 (Quartile: 492, 1567) respectively. 50% of the reviews were last assessed as up-to-date prior to 2008, of reviews considered out-of-date, one was last updated in 2000. In total, 44 diseases were investigated in the included reviews, 18 (31%) reviews addressed cerebral vascular and cardiovascular diseases (9 reviews focused on stroke), followed by reviews focused on respiratory diseases (n = 6) and gynecological/pregnancy diseases (n = 6). Of the 51 nonempty reviews, only one review concluded positively, 27 (53%) concluded that there might be benefit of CHM for treating specific health conditions, which was limited by the poor quality or inadequate quantity of studies, 23 (45%) reviews concluded that the currently available data do not allow any conclusion to be drawn, generally because of low methodological quality of studies, small number of studies and participants included or publication bias. Nineteen reviews focused on 13 single herbs or herbal preparations, while the remainder (39 [67%]) addressed broad questions, in which 34 reviews concerned multiple Chinese herbs or multiple formulations of Chinese herbs, 5 reviews involved a family of herbal medicines sharing similar efficacy, including Huangqi type formulations (including Huangqi injection and Huangqi-Danggui mixture), Chuanxiong preparations (including Nao-an capsule, Xifeng wan and Apoplexy Preventing Dry Ointment Powder), Dan Shen agents (including Compound Danshen Dripping Pill, Compound Danshen injection, Danshen injection, Yiqi huoxue injection, and Quyu huatan xiezhuo fang) and Sanchi (including Xinnaotai, Sanchitongshu capsule, Naoming injection, Xuesaitong soft capsule, Sanqitongshu capsule, Xuesaitong and Xueshuantong injection). Of the 39 reviews, 4 didn't include any primary study, of the remaining reviews (n = 35), the median number of herbal medicines involved was 6 within a range of 1 to 71, results of different herbal medicines were pooled in 9 reviews, in which 7 reviews pooled the results of all Chinese herbal medicines, one review pooled all Danshen agents, and one pooled Sanchi. Table 1 presents a summary of OQAQ items of the included reviews, the mean score (item 10) was 5.05, 95% CI (4.58, 5.52). 41 of 58 reviews attempted to minimize bias during the selection of studies by at least two reviewers independently select eligible studies. All reviews reported assessing the methodological quality of included primary studies, 21 (36%) reviews used the Cochrane Collaboration's 'Risk of Bias' tool, the Jadad scale was used in 6 reviews, 12 (21%) reviews used unnamed checklist. Among 671 included studies, 108 (16%) used adequate sequence generation, allocation concealment was adequate in 50 (7%) studies. The median number of databases searched in 58 reviews was 7 within a range of 4 to 15. Regarding to the English language databases, the most searched was MEDLINE (98%), followed by EMBASE (97%) and CENTRAL (97%). CBM was the most searched Chinese database (78%), the second most used was CNKI (45%) and the third was VIP (24%) ( Table 2 ). All reviews searched at least 2 English databases, while 10 reviews did not search any Chinese database. 41 (71%) reviews searched at least 3 English databases, while only 6 (10%) reviews searched at least 3 Chinese databases (Table 3 ). Evidence-based health care involves the systematic collection, synthesis and application of scientific evidence to guide clinical practice and policy-making. Systematic reviews are a key component of evidence-based health care. Currently, CDSR has included 58 systematic reviews of CHM. Almost seventy percent of included reviews' topics were too broad, the percentage was much higher than that of similar reviews published in Chinese journals, which was 38 percent (41 among 107 reviews) [69] . It is difficult to develop a comprehensive search strategy for broadly focused reviews, for instance, one hundred and sixty herbal medicines are now available for coronary heart diseases treatment, a systematic review of CHM for coronary heart disease will have to include clinical trials of all these herbal medicines, it is easily to cause the incomplete identification of relevant studies. Choosing broad topics for reviews will require more resources in data collection and analysis, the results may also be too complicated to interpret. So topic selection of CHM reviews should focus on specific clinical problems, the extensive titles are not recommended. As broad questions of reviews may be addressed by large sets of heterogeneous studies, the data synthesis may be particularly challenging. In our study, 9 reviews (among the 39 broadly focused reviews) pooled the results of different herbs, which did not identify potentially important differences in effects across different interventions. Systematic reviews can, but do not have to use meta-analysis when combining data from primary studies, prior to conducting a meta-analysis, reviewers should examine the consistency of the interventions. It is recommended that the data of each intervention should be analyzed and presented separately if several different interventions for the same condition were tested in one review. The goal of a systematic review is to identify relevant studies completely and unbiasedly [70] . It has been demonstrated [71, 72] that significant amounts of evidence would potentially be missed if the search is limited to English-only sources. Because a considerable number of clinical trials on CHM were published only in the Chinese language journals, so a comprehensive search of Chinese databases is essential for a systematic review of CHM. However, we were disappointed to find that almost twenty percent of reviews did not search any Chinese database in our study. One study [73] compared four Chinese databases and concluded that CBM is the preferred database for systematic reviewers to retrieve relevant Chinese studies, while CNKI is recommended for non-Chinese-speaking researchers due to its free searched English version website (www.global.cnki.net) and ''Cross-Language Search'' functions. CBM has no English website and a fee is charged for searching, now many Chinese medical universities have got the permission to search CBM, so maybe the most costand time-efficient way to search it is to enhance collaboration with Chinese researchers. In doing a review of CHM, professional advice from TCM practitioners is of great value. It is generally assumed that the characteristics of TCM would be well taken into account in a review if one or some of its reviewers majored in TCM. In our study, we found that more than sixty percent of reviews did not have one TCM practitioner in the authors list, which might lead to insufficient consideration of the characteristics of TCM (e.g. determination of treatment based on pathogenesis obtained through differentiation of symptoms and signs) and incorrect results. We suggest that future reviews should be authored by a group of individuals with both clinical expertise and methodological expertise. Of the 671 primary studies included, less than twenty percent of studies used adequate sequence generation, and only seven percent used adequate allocation concealment. Because of the poor methodological quality of primary studies, nearly half of reviews were reported as being inconclusive, while 27 reviews provided preliminary evidence of CHM's benefits to certain conditions, which should be considered tentative and need to be confirmed with rigorous RCTs. The Chinese government has been aware of the importance of conducting scientifically sound RCTs and has made substantial investments into funding clinical researches of CHM, now many well-designed RCTs of CHM with rigorous methodology are in progress or have been completed in China [74] , we believe future updates of currently inconclusive Cochrane reviews of CHM may reach more definitive conclusions. Although we believed a review has a greater chance of considering the characteristics of TCM if at least one author works in TCM department, university or hospital, or it stated that it had got suggestion from TCM practitioners, it is quite possible that some reviewers had consulted TCM experts while designing and doing the review, but did not report it in articles. As cases of this kind could not be ruled out, we therefore might have underestimated the proportion of reviews getting support from TCM practitioners. In addition, we restricted our search to Cochrane reviews because they are generally less prone to bias than systematic reviews published in paper-based journals [9, 75, 76] . However, this might cause the results of this study to be only applicable to review articles in the Cochrane database. Further evaluation is needed in order to know whether the systematic reviews of CHM published in English paper-based journals even in the leading journals have the same problem. Table S1 List of all included Cochrane reviews of CHM. (DOC) PRISMA Checklist S1 (DOC) PRISMA Flow Diagram S1 (DOC)

Current Status of the Immunomodulation and Immunomediated Therapeutic Strategies for Multiple Sclerosis

Multiple sclerosis (MS) is an autoimmune disease of the central nervous system, and CD4(+) T cells form the core immunopathogenic cascade leading to chronic inflammation. Traditionally, Th1 cells (interferon-γ-producing CD4(+) T cells) driven by interleukin 12 (IL12) were considered to be the encephalitogenic T cells in MS and experimental autoimmune encephalomyelitis (EAE), an animal model of MS. Currently, Th17 cells (Il17-producing CD4(+) T cells) are considered to play a fundamental role in the immunopathogenesis of EAE. This paper highlights the growing evidence that Th17 cells play the core role in the complex adaptive immunity of EAE/MS and discusses the roles of the associated immune cells and cytokines. These constitute the modern immunological basis for the development of novel clinical and preclinical immunomodulatory therapies for MS discussed in this paper.

Multiple sclerosis (MS) was initially identified in 1868 by Charcot. This disease often begins in young adulthood with intermittent episodes of neurological dysfunction, including visual impairment, ataxia, motor and sensory deficits, and bowel and bladder incontinence. These are attributable to recurrent inflammatory attacks on the white matter of the brain and spinal cord, which lead to the accumulation of perivascularly distributed inflammatory cells within the brain and spinal cord white matter [1] . Beeton et al. first established an animal model of MS in the 1930s, when they immunized monkeys with a central nervous system (CNS) homogenate to induce what is now known as experimental autoimmune encephalomyelitis (EAE) [2] . Since this pilot animal study, EAE has become the most accepted animal model of MS. In recent decades, pathogenic hypotheses have been investigated and novel therapeutic agents tested in this model in the fields of CNS inflammation and demyelination. Therefore, EAE provides a valuable tool for the investigation of the T-cell-dependent pathogenesis of autoimmune inflammation in the CNS and the orchestration of the autoimmune demyelinating inflammation in the CNS of MS patients. Mice and/or genetically modified mice have also been of fundamental value in the exploration of the complex pathogenesis of MS [3, 4] . EAE is undoubtedly the best animal model in which to study autoimmune diseases and particularly the demyelinating diseases of the CNS, such as MS [5] . Myelin basic protein-(MBP)-specific T cells isolated from the peripheral lymphocytes of human individuals with MS and encephalitogenic T cells recovered from circulating 2 Clinical and Developmental Immunology autoreactive T cells of either immunized or naïve animals have shown that autoreactive T-cell lines that recognize the encephalitogenic part of MBP in vitro can be distinguished from an unprimed rat T-cell population. This confirms that autoreactive T cells play a central role in the pathology of MS [6] [7] [8] . EAE can also be induced by adoptively transferring an expanded population of myelin-reactive encephalitogenic CD4 + (T helper [Th] ) cells, which allows the further dissection of the immunopathogenic potency of different encephalitogenic CD4 + cell populations [9] . In the 1990s, Mosmann and Coffman postulated that Th cells can be classified into two distinct subsets, Th1 and Th2. Th1 cells produce large quantities of interferon γ (IFNγ), driven by interleukin 12 (IL12), which promotes cellular immunity directed against intracellular pathogens. Alternatively, Th2 cells, which secrete IL4, IL5, IL13, and IL25, are essential in the destruction of extracellular parasites and the mediation of humoral immunity [10, 11] . Selfreactive Th1 clones derived in vitro are capable of adoptively transferring EAE to naïve recipients [12] . Increased levels of Th1 cytokines are particularly evident during EAE/MS relapse, whereas increased Th2 cytokines are found during remission in MS patients when compared with control levels [13] . Clinical and hematological symptoms are exacerbated in relapsing/remitting MS patients following the administration of IFNγ, and this is also observed in other Th1type diseases, whereas it is less apparent in Th2 diseases [14, 15] . Th1 cells were earlier thought to be pathogenic T cells, whereas Th2 cells were thought to confer an antiinflammatory potential, constituting protective T cells in both MS and EAE [16] [17] [18] [19] . However, this clear-cut immunodysregulation of the Th1/Th2 balance in EAE and MS may be part of a hidden complex of interactions underlying EAE and MS [20] . The Th1-driven nature of the EAE/MS disease was challenged by the finding that IFNγ-and IFNγ-receptor-deficient mice, as well as mice that lack other molecules involved in Th1 differentiation, such as IL12p35, IL12 receptor β2 (IL12Rβ2), and IL18, were not protected from EAE, but instead were more susceptible to the disease [21] [22] [23] [24] [25] . Unexpectedly, mice deficient in IL12α (IL12p35), a component of the Th1 paradigm, are vulnerable to EAE. Similarly, IL12Rβ2deficient mice develop more severe clinical manifestations of EAE, whereas IL12p40-deficient mice are resistant to EAE [23, 24, 26] . These discrepancies and conflicting data indicate that an imbalance in the Th1/Th2 milieu cannot explain the overall immunopathogenic mechanisms underlying EAE and MS. p19, a novel cytokine heavy-chain homologue of the IL6 subfamily, was discovered as a computational sequence [27] . When the p19 chain is linked to the p40 chain, a subunit of IL12 (another subunit of the IL12 heterodimers is the p35 chain), it forms a novel cytokine designated IL23. Therefore, the deletion of IL12p40 will affect the functions of both IL12 and IL23. Cua and colleagues verified that Il23 but not Il12 is essential for the induction of EAE by generating Il23p19 knockout (KO) mice and comparing them with IL12p35 KO mice [28] . Furthermore, an IL17-producing Tcell subset, driven and expanded by IL23, can pathogenically induce EAE when adoptively transferred into naïve wildtype mice [29, 30] . These IL17-producing T cells were dramatically reduced in the CNS of IL23p19-deficient mice. Based on these studies, researchers confidently suggested that IL17-producing CD4 + T cells are a distinct and novel Th subset that exacerbates autoimmunity, and designated them Th17 cells [31, 32] . Th17 cells are a Th-cell subset distinct from Th1 and Th2 cells in terms of their differentiation, expansion, and effector functions [33, 34] . The discovery of Th17 cells further clarifies the cytokine profile of MS [35] . Recently, the levels of IL17 produced by MBP-stimulated peripheral blood cells obtained from MS patients or controls were shown to correlate with the active lesions in MS patients observed with magnetic resonance imaging (MRI) [36] . Like other Th subsets, the Th17 lineage is activated by a specific cytokine milieu. However, IL23 cannot produce Th 17 cells de novo from naïve T cells, and the IL23 receptor (IL23R) is not expressed on naïve T cells [37] . Transforming growth factor β (TGFβ) upregulates IL23R expression, thereby conferring responsiveness to IL23, which confirms that TGFβ is a critical cytokine in the commitment to Th17 expansion in vitro and in vivo [38] . In mice, TGFβ together with IL6 can activate antigen-responsive naïve CD4 + T cells to develop into Th17 cells [39] . In humans, naïve CD4 + cells exposed to IL6, TGFβ, and IL21 can develop into Th17 cells, and the production of IL23 plays a role in maintaining these Th17 cells [40, 41] . Altogether, Th17 cells require IL23, TGFβ, IL6, and IL1 for their generation. Th17 cells produce IL17A and IL17F, which are upregulated in chronic lesions [42] , and IL22, which is also involved in the pathogenesis of MS. Thus, Th17 cells are a recently discovered, unique Th lineage that produces a repertoire of signature cytokines, including IL17A, IL17F, IL21, and IL22, that are essential for the development of autoimmune diseases such as MS [43] . The discovery of transcription factors that are key regulators of the cytokine expression required to launch lineagespecific transcriptional programs has greatly extended our understanding of Th-cell lineage commitment [44] . It has been shown that T-bet and STAT4 program the commitment of the Th1 lineage and Th1 cytokine production [45] , whereas GATA-binding protein 3 (GATA3) and STAT6 drive Th2 population expansion and Th2 cytokine production [46, 47] . The T-bet and STAT4 (necessary for Th1 differentiation) transcription factors are important in the differentiation of autoimmune T cells in the EAE model [48] , and T-betand STAT4-deficient mice are resistant to EAE. However, these transcription factors do not mediate the induction of Th17 cells. Instead, in a unique inductive milieu, Th17 differentiation is driven by distinct transcription factors: retinoic acid receptor-related orphan receptor-γt (Rorγt) and Rorα [33, 34] . Stat3 deletion in T cells also prevents autoimmune uveitis and EAE and increases the expression of IL10 and forkhead box P3 (FoxP3) [49] , and the expression of FoxP3 programs the development and functions of Treg cells [50] . In humans, IL23 and IL1β also induce the development of Th17 cells expressing IL17A, IL17F, IL22, IL26, IFNγ, the chemokine CCL20, and the transcription factor RORγ [51] [52] [53] , as illustrated in Figure 1 (adapted from Hirota et al. [54] ). Recent microarray studies of lesions in MS patients demonstrated an increased expression of IL17, confirming that Th17 cells play an important role in the development of inflammation and demyelination and in the eventual damage of the CNS. IL17 is a recently described cytokine produced in humans almost exclusively by activated memory T cells and can induce the production of proinflammatory cytokines and chemokines from parenchymal cells and macrophages. Patients with MS have greater numbers of IL17-mRNAexpressing mononuclear cells in the cerebrospinal fluid (CSF) than in the blood. Previously, no increase in the numbers and expression of IL17 mRNA by mononuclear cells isolated from the CSF was observed in patients with MS, but higher levels of IL17 mRNA were observed in the CSF than in the blood, with the highest levels in the blood detected during clinical exacerbations [56] . These data confirm the pivotal role of IL17 in MS both peripherally and centrally. Myelin is expressed in the circulation, and other CNS antigens are thought to be expressed in the cervical lymph nodes, which can trigger the conversion of autoaggressive myelinreactive T cells to pathogenic T cells. Adhesion molecules, the integrins, allow these myelin-reactive T cells to penetrate the blood-brain barrier (BBB) under inflammatory conditions, and in this way, activated and memory T cells can enter the CNS [57] . Autoaggressive myelin-reactive T cells migrate into the CNS, where they recognize their cognate target antigens, and the movement of antigen-presenting cells (APCs) into the CNS is essential for lymphocyte reactivation within the CNS compartment and the initiation of the inflammatory cascade in the development of EAE [58] . Subsequently, inflammatory and immune cells, such as granulocytes and macrophages, are attracted into the CNS parenchyma, where they mediate tissue inflammation, leading to demyelination and tissue damage [59] . The brain was formerly considered an immunoprivileged organ, but this perspective has been revised in the last two decades [60] . Today, we understand that any damage to the CNS can activate immune cells in situ in the CNS, particularly microglial cells. Deshpande et al. demonstrated the transient inactivation of microglial cells via a cell-specific deficiency of CD40 expression, indicating that microglial cells are crucial for maintaining the autoimmune responses in the CNS [61] . The major histocompatibility complex (MHC, also known as "human leukocyte antigens" in humans) class II molecules are only displayed on specialized APCs (e.g., dendritic cells [DCs], B cells, and macrophages), whereas MHC class I molecules are expressed by all cells in the inflammatory milieu of the CNS [62] . Microglial cells upregulate the expression of MHC and costimulatory molecules to initiate the generation and maintenance of the inflammatory milieu. DCs seem to play a critical role in antigen presentation to invading T cells and in the release of cytokines and chemokines, thereby guiding the entry of monocytes, lymphocytes, and cells with a phenotype similar to that of DCs into the lesion [63] . Th cells recruit macrophages, which release proinflammatory cytokines and destructive molecules (such as nitric oxide [NO], IL1, IL6, tumor necrosis factor α [TNFα], and matrix metalloproteinases (MMPs]), and CD8 + T cells also directly attack MHC class I-expressing cells, such as oligodendrocytes and neurons [64, 65] . The secretion of destructive molecules, such as NO and TNFα, and the degradation of myelin are consequences of this cascade. TNF receptor 1 (TNFR1) but not TNFR2 signaling is critical for demyelination and the limitation of T-cell responses during immune-mediated CNS disease [66] . This complicated process triggers the recruitment of innate immune cells, generally consisting of T cells, macrophages, and microglia, which in turn mediate demyelination, axonal damage, and lesions. In autopsy samples from MS patients, the expression of IL17 is evident in perivascular lymphocytes and in astrocytes and oligodendrocytes located in the active areas of CNS lesions. IL17R is also identifiable in acute and chronic MS plaques of patients with MS, suggesting the enrichment of Th17 and CD8 + T cells in active MS lesions, and confirming an important role for IL17 in the pathogenesis of MS [67] . Th17 cells are identified by their expression of IL23R and the memory T-cell marker CD45RO in situ. Other markers that have been investigated including the chemokine receptor, CCR6, and RORC variant 2, which is a central transcription factor for Th17-cell development [42, 68] . Microarray analysis of MS lesions has also demonstrated increased transcripts of genes encoding inflammatory cytokines, particularly IL6, IL17, and IFNγ and associated downstream pathways [56] . A significant increase in IL23 mRNA and protein expression is found in lesion tissues compared with nonlesion tissues. Activated macrophages/microglia have been shown to be important sources of IL23p19 in active and chronically active MS lesions. IL23p19-expressing mature DCs are preferentially located in the perivascular cuffs of active lesions. This data on the expression of IL23p19 in MS lesions improves our understanding of the pathogenesis of MS [69] . There is also evidence that MS endothelial cells express high levels of IL17R and are more permeable to IL17 than are non-MS endothelial cells. Perivascular DCs also express high levels of granzyme B in inflammatory lesions, polarizing naïve CD4 + T cells into Th17 cells. These Th17 cells transmigrate efficiently across BBB endothelial cells (BBB-ECs), leading to the destruction of human neurons and initiating CNS inflammation through Th-cell recruitment [70] . Similarly, the expression of IL17R and IL22R on BBB-ECs has been examined in MS lesions, and IL17 and IL22 have been shown to disrupt BBB tight junctions in vitro and When naïve CD4 + TCRαβ + T lymphocytes, classified by their low expression of CD44, absence of CD25, and high levels of CD62L, encounter their cognate antigens, they can differentiate into several previously identified effector subsets. It is likely that several "master" transcription factors, individually required for T-cell differentiation towards one of the end effector stages, are initially expressed upon engagement of the TCR with costimulatory receptors. Each transcription factor drives a specific set of genes required for lineage commitment and the expression of signature cytokines and negatively affects alternative pathways. However, the local microenvironment is the driving force that determines the outcome of the differentiation course. Th1 cells are established in the presence of IFNγ and IL12 and signaling via STAT1 and STAT4, resulting in the expression of the master transcription factor T bet. Th2 cells depend on IL4 and STAT6 for the increased expression of GATA3, whereas the simultaneous presence of TGFβ results in the development of Th9 cells, utilizing an undefined master transcription factor. The presence of TGFβ, with IL2 signaling via STAT5, is known to generate, at least in vitro, inducible Treg, which utilize FOXP3 like those Treg generated in the thymus. Again, it is TGFβ in combination with IL6 signaling via STAT3 that drives the expression of RORγt, resulting in the differentiation of Th17 cells. However, the initiation of the developmental program of these T helper subsets may not be completed in the presence of only these driving cytokines. Several additional factors may be required for their subsequent functional maturation or may be responsible for the fine tuning of their effector phases. Several of these factors are indicated, together with the characteristic cytokine profiles of each subset (adapted from [54] ). in vivo. IL6 transsignaling may also play a role in the autoimmune inflammation of the CNS, mainly by regulating the early expression of adhesion molecules, possibly via cellular networks at the BBB [71] . Ifergan et al. demonstrated that a subset of CD14 + monocytes migrate across the inflamed human BBB and differentiate into CD83 + CD209 + DCs under the influence of BBB-secreted TGFβ and granulocytemacrophage colony-stimulating factor (GM-CSF). These DCs can produce IL12p70, TGFβ, and IL6 and promote the proliferation and expansion of distinct populations of Th1 and Th17 cells. The abundance of such DCs in situ is strongly associated with microvascular BBB-ECs within acute MS lesions and with a significant number of Th17 cells in the perivascular infiltrate [72] . Astrocytes play significant physiological roles in CNS homeostasis and act as a bridge between the CNS and the immune system. Astrocytes also contribute to the complex interactions during CNS inflammation. IL17 functions in a synergistic manner with IL6 to induce IL6 expression in astrocytes. Astrocytes upregulate the expression of IL17 and IFNγ genes and proteins in T cells, which is consistent with the astrocytes' capacity to express IL23 subunit p19 and the common IL12/IL23 subunit p40, but not IL12 subunit p35 when these two cell types are cocultured [73] . Das Sarma et al. demonstrated increased IL17RA expression in the CNS of mice with EAE and the constitutive expression of functional IL17RA in mouse CNS tissues. They also identified the expression of IL17RA in both astrocytes and microglia in vitro. In that study, the secretion of the chemokines Mcp1, Mcp5, Mip2, and CxcL1 was upregulated in these cells, suggesting that the upregulation of chemokines by glial cells is the result of IL17A signaling through constitutively expressed IL17RA [74] . Ma et al. demonstrated that the suppressor of cytokine signaling 3 (Socs3) participates in IL17 functions in the CNS as a negative feedback regulator, using mouse models of Socs3 small interfering RNA (siRNA) knockdown and Socs3 deletion. These mice with loss of Socs3 function showed enhanced IL17 and IL6 signaling in astrocytes via the activation of the NF-κB and Mapk pathways, indicating that astrocytes can act as a target of Th17 cells and IL17 in the CNS [75] . Similarly, Kang et al. constructed specific deletion mutants of Act1, a critical component required for IL17 signaling, in mice with EAE to examine CNS inflammation in endothelial cells, macrophages, microglia, and the neuroectoderm (neurons, astrocytes, and oligodendrocytes). In these Act1-deficient mice, Th17 cells showed normal infiltration into the CNS but failed to recruit lymphocytes, neutrophils, and macrophages. Therefore, astrocytes are critical in IL17-Act1-mediated leukocyte recruitment during EAE [76] . Interestingly, Merkler et al. demonstrated that macrophages respond to the Th1 milieu and neutrophils respond to Th17 cytokines in a marmoset monkey model of EAE. They also showed dense accumulations of T and B lymphocytes, MHC-II-expressing macrophages/microglia, and early activated macrophages at the sites of perivascular and parenchymal lesions in the neocortex and subcortical white matter, indicating that the inflammatory response, especially macrophage and microglia activation, may be regulated differently in the gray matter areas of the primate brain [77] . In summary, DCs in the peripheral tissues and microglia in the CNS are responsible for cytokine polarization and the expansion of Th17 cells. The complex interactions of Th17 cells with different DCs, such as microglia, astrocytes, and peripheral DCs (including neutrophils and macrophages), all contribute to the immunopathogenesis of EAE and MS. IL1R KO mice have impaired Th17 cells and are protected from EAE [78] , and IL1β increases the susceptibility to and progression of relapse onset in MS [79] , implying a role for IL1β in the development of EAE and MS. EAE was abolished by a virus-expressing IL4 but not by a virus-expressing IL10 in chronic relapsing EAE. Therefore, the cytokine environment was converted from a disease-promoting IL23producing condition to a disease-limiting IL4-producing condition by the local expression of IL4 from a Herpes simplex virus vector delivered to the brain [80] . Moreover, the increased expression of IL4 in glial cells was associated with the reduced severity of EAE [81] , suggesting that the upregulation of Th2 cytokines inhibits the propagation of the inflammation of EAE/MS by encephalitogenic Th17 cells. CD4 + CD25 + Foxp3 + T cells, well-known regulatory T cells (Tregs), retain the potential to inhibit the autoimmune response, and protect against inflammatory injury. TGFβ is a key cytokine in the generation of Tregs. Tregs are not only primarily involved in the regulation of Th17 cells but can also regulate the functions of Th1/Th2 cells [82] . A distinction has been drawn between the generation of pathogenic Th17 cells that induce autoimmunity and the generation of Tregs that inhibit autoimmune tissue injury [39] . Although EAE was once considered a classical Th1 disease, it has been proposed that it is predominantly Th17 driven. Recently, Singh et al. demonstrated that the overexpression of IL17 in T cells did not exacerbate EAE. Moreover, genetic and antibody studies have indicated that the absence of IL17A or IL17F does not reduce the incidence or severity of EAE. The collective findings of IL17 and IFNγ studies indicate that their roles may depend on the nature of the immune response and that the IL17 that occurs in the brain may overcome the inhibitory effect of IFNγ, which generally prevents inflammation at that site [83] . When pure Th17 cells from myelin oligodendrocyte glycoprotein-(MOG-) immunized mice, polarized with TGFβ to deplete any IFNγ production, are adoptively transferred to mice, they do not induce EAE, suggesting that the reciprocal interactions among Th17-related cytokines enrol and activate the involvement of associated immune cells. Interestingly, when Th17 cells are combined with Th1 cells, they can fully induce EAE disease [84] . Liu et al. also demonstrated that the loss of STAT3 by Th cells results in an intrinsic developmental defect that renders STAT3 −/− mice resistant to CNS inflammatory diseases. STAT3 is required for the production of IL17 by Th17 cells, the generation of double positive T cells expressing IL17 and IFNγ, and T cell trafficking into CNS tissues. This suggests that STAT3 may be a therapeutic target for modulating CNS autoimmune diseases, and that Th1 cells can facilitate the entrance of Th17 cells into the CNS during EAE [85] . An encephalitogenic Th1 cell line that induces the recruitment of host Th17 cells to the CNS during the initiation of EAE has been reported [49] . Stromnes et al. showed significant differences in the regulation of inflammation in the brain and spinal cord, depending on different Th17/Th1 ratios, by demonstrating that specific T-cell populations targeting different myelin epitopes are characterized by different Th17/Th1 ratios in EAE [86] . Therefore, Th1 cells have the potential to reciprocally regulate Th17 cells during EAE. IL21 is a type I four-α-helix bundle cytokine that belongs to the IL2 family and functions as a "growth hormone"like cytokine. After the antigen-responsive differentiation phase, Th17 cells enter the amplification stage, and IL21 plays a pivotal role in the expansion and differentiation of the Th17 lineage, providing an autocrine and paracrine stimulus for Th17 cells [41, 87] . During clonal expansion, IL21 also promotes IL23R expression in differentiated Th17 cells, which plays an important role in the stabilization of the Th17 lineage in the presence of IL23 [88] . Although no effects were observed when Il21 was administered after EAE progression, the administration of IL21 boosted natural killer (NK) cell functions before the induction of EAE, including the secretion of Ifnγ. Therefore, IL21, by affecting NK cells, has various effects during the initiation and progression of EAE [89] . Alternatively, IL27, an IL12/IL23 family member, is a negative regulator of Th17 cell differentiation and can prevent inflammatory demyelination in the EAE model [44] . IL27 drives the expansion and differentiation of IL10producing Tr1 cells by inducing the expression of three key molecules: the transcription factor c-MAF, the cytokine IL21, and ICOS. Moreover, IL27-driven c-MAF expression transactivates the production of IL21, which acts as an autocrine growth factor for the expansion and/or maintenance of IL27-induced Tr1 cells. ICOS also promotes IL27-driven Tr1 cells. Each of these elements is essential, because the loss of c-MAF, IL21 signaling, or ICOS reduces the frequency of IL27-induced differentiation of Tr1 cells ( Figure 1 ) [90] . Exacerbation of EAE was demonstrated in IL27-deficient mice, and interestingly, Il27-treated mice had markedly reduced CNS inflammatory infiltration, indicating the downregulation of Th17 phenomena [91] . Recently, a novel effector T-cell subset, Th9 cells, has been identified, and the ability of this T-cell subset to induce EAE is currently being investigated. Jäger et al. generated Mog-specific Th17, Th1, Th2, and Th9 cells in vitro to directly characterize their encephalitogenic potency after their adoptive transfer. They found that Mog-specific Th1, Th17, and Th9 cells, but not Th2 cells, induce EAE. Interestingly, each T-cell subset induced disease in a distinct pathological manner, suggesting that the different effector Th subsets that induce EAE do so differently and implying that the pathological heterogeneity in MS lesions might be partly attributable to various characteristics of myelinreactive effector T cells [92] . The authors also suggested that MS might be a disease caused by multiple distinct myelinreactive effector cells. The disease induced by Th17 cells in some animals exhibited symptoms atypical of EAE, including ataxia, severe imbalance, and weight loss associated with high mortality. Some animals had a mixture of atypical and typical EAE symptoms. When cells were recovered from the CNS, it appeared that the transferred Th9 cells produced IFNγ. The identities of the other cell populations did not seem to drift after their in vivo transfer [93] . Nowak et al. recently demonstrated that like other T cells cultured in the presence of TGFβ, Th17 cells produce IL9. Th17 cells generated in vitro with IL6 and TGFβ and ex vivo-purified Th17 cells both produced IL9. Data show that IL9 neutralization and IL9R deficiency attenuate the disease, and this correlated with reductions in Th17 cells and IL6-producing macrophages in the CNS. These authors also confirmed the role of IL9 in the development and progression of EAE and implicated Il9 as a Th17derived cytokine that contributes to inflammatory disease [94] . Together, Th2 cells, Tr1 cells, and Tregs exert repressive effects on Th17 cells, and Th9 cells have a stimulatory effect on Th17 cells, suppressing EAE and MS. However, Th1 cells play dual roles in EAE. Our understanding of the pathophysiology and neurodegenerative processes of MS has led to the development of novel therapeutic strategies. Since the early 1990s, diseasemodifying drugs have been introduced for the selective management of MS, including IFNβ and glatiramer acetate (GA), which have become the standard treatment for relapsing/remitting MS [95] . Most recommendations previously made by the Multiple Sclerosis Therapy Consensus Group (MSTCG) on the use of disease-modifying drug therapies remain valid [96, 97] . Hermmer and Hartung have published an apparent review of the development of rational therapies in MS [98] . Therefore, we will discuss four domains of novel immunomediated therapeutics used for MS and their current status. The first domain includes immunosuppressive agents, such as mitoxantrone, laquinimod (ABR-215062), cladribine (Mylinax ), and teriflunomide (probably via the suppression of TNFα and IL2 production). The second domain includes immunomodulatory agents: (1) cytokine inhibitors such as IFNβ; (2) agents that deplete specific immune cell subsets, such as alemtuzumab (a human monoclonal antibody [mAb] that targets CD52 expressed by T and B cells, producing long-term T-cell depletion) [99, 100] and rituximab (which targets CD20 to deplete human B cells) [99, 101] ; (3) agents that selectively block coreceptors and costimulators, such as daclizumab (an anti-CD25 mAb that inhibits activated T cells and induces regulatory immune cells) [102] . The third domain involves the development of migration-modifying therapies: (1) agents that affect adhesion molecules, such as natalizumab (an mAb that blocks very late antigen 4 [VLA-4]) and (2) sphingosine 1phosphate receptor (S1PR) agonists: fingolimod (FTY720). The fourth domain includes neuroprotective agents associated with immunomodulation, including broad-spectrum immunomodulators such as statins, PPAR agonists (e.g., pioglitazone, gemfibrozil), the sex hormone estriol (E3), fumarate, minocycline, and erythropoietin (EPO), all of which have been effective in the treatment of both EAE, and MS. IFNβ has been clinically introduced to treat patients with MS based on its ability to shift a Th1-mediated response to a Th2-mediated response [92] . However, microarray studies have indicated that a number of genes in patients with MS are upregulated by the cytokines associated with the differentiation of cells into Th1 lymphocytes rather than into Th2 lymphocytes, suggesting that this shift may not be the only therapeutic mechanism of IFNβ in MS [103] . IFNβ therapy also reduces IL23 mRNA levels [104] . IFNβ inhibits human Th17 cell differentiation, so the Th17 axis could be another target of IFNβ therapy [105] . IFNβ-mediated IL27 production by innate immune cells has been shown to play a critical role in the immunoregulatory role of IFNβ in EAE by inhibiting Th17 cells in EAE mice and MS patients [91, 106, 107] . Besides, Galligan et al. evidence further that IFNβ(−/−) mice exhibited an earlier disease onset and a more rapid progression of EAE compared to IFNβ(+/+) mice of EAE and IFNβ(−/−) mice of EAE had increased 7 numbers of CD11b(+) leukocytes infiltrating affected brains and an increased percentage of Th17 cells in the CNS with augmentation of autoreactive T cells,suggesting that IFNβ acts to suppress the production of autoimmune-inducing Th17 cells during the development of disease as well as modulating proinflammatory [108] . In addition, the therapeutic effect of IFNβ is probably attributable to the induction of the regulatory cytokine IL10 [104] . Furthermore, Axtell et al. design a delicate study to further clarify the role of IFNβ in MS/EAE [109] . Likewise, They demonstrate that IFNβ was effective in reducing EAE symptoms transferred by Th1 cells transfer but exacerbated disease by Th17 cells transfer and effective treatment of IFNβ in Th1-induced EAE correlated with augmented IL10 production; differently, in Th17-induced EAE, the amount of IL10 was unaffected by treatment of IFNβ. Likewise, a high IL17F level in the serum of people with RRMS is associated with fail of IFNβ therapy. This characteristic of IFNβ might contribute to explore some logical biomarkers for predictive assessment of the response to a popular therapy for MS [109, 110] . Although, B cells may have a dual role in the pathogenesis of MS that they contribute to the induction of the autoimmune response but also mediate the resolution of the CNS inflammatory infiltrate [111, 112] . However, Ramgolam et al. demonstrate further that supernatants transferred from IFNβ-1b-treated B cells inhibited Th17 cell differentiation, as they suppressed gene expression of the RORC and IL-17A and secretion of IL-17A. Likewise, IFNβ-1b also induces B cells' IL-10 secretion which may mediate their regulatory potent [113] . Thus, IFNβ-1b exerts its therapeutic effects at least in part by targeting B cells' functions that contribute to the autoimmune pathogenesis of RR MS, which may uncover extra mechanisms of the B-cell contribution to the autoimmune effects and provide novel targets for future selective treatment of MS [113] . Glatiramer acetate (GA; Copaxone; copolymer 1) exerts a clinical response in MS patients via its modulation of IFNγ and IL4 by reducing the expression of IFNγ and ensuring the stable expression of IL4 in anti-CD3/CD28stimulated peripheral blood mononuclear cells (PBMCs) [114] . Moreover, GA enhances the suppressive effects of Tregs in both EAE and MS [115, 116] . Studies of human DCs have shown that GA modulates the production of inflammatory mediators without affecting DC maturation or immunostimulatory potential. DCs exposed to GA secrete low levels of the Th1-polarizing factor IL12p70 in response to lipopolysaccharide and triggering of the CD40 ligand [117] . Human DCs exposed to GA also induce IL4-secreting effector Th2 cells and increase their expression of IL10 [118] . These results show that APCs, including DCs, are essential for the GA-mediated shift in Th-cell phenotypes and indicate that DCs are an important target of the immunomodulatory effects of GA. Patients with MS show a threefold to fourfold increase in the expression of the α4 subunit of the integrin VLA-4, which is normally expressed on activated lymphocytes, monocytes, and other cell types in the CSF and circulation [119] . Elovaara et al. confirmed that methylprednisolone reduces the adhesion molecules in the blood and CSF in patients with MS [120] , implying that targeting leukocyte trafficking may be a possible therapeutic strategy for MS [121] . Therefore, natalizumab, a humanized mAb directed against the VLA-4 adhesion complex, has been introduced into the treatment of MS and reduces the risk of sustained progression of disability and the rate of clinical relapse in patients with relapsing MS [122] . However, during clinical trials, two natalizumab-treated MS patients developed progressive multifocal leukoencephalopathy (PML), which resulted in the voluntary removal of the drug from the market in February 2005 [123, 124] . A retrospective safety evaluation was subsequently conducted, and natalizumab was consequently returned to the market as a monotherapy in July 2006 for the treatment of relapsing MS; however, there were 111 cases of PML reported subsequently in natalizumab-treated MS patients as of April 2011 [125] . More evidently, the risk of developing PML for a MS patient on natalizumab (Tysabri) is almost 100 times higher if the patient (1) has been taking the drug for more than two years, (2) has a prior history of immunosuppressant use, and (3) tests positive for antibodies to the JC virus [126] , compared to a patient with none of these three risk factors [127] . Instead, there is currently no convincing evidence that natalizumab-associated PML is restricted to combination therapy with other disease-modifying or immunosuppressive agents [128] . Nevertheless, natalizumab use must be restricted to the indicated patients. Mitoxantrone, a cytotoxic drug with immunomodulatory properties, is used to treat progressive forms of MS [129] . Mitoxantrone increases the ex vivo production of the Th2 cytokines IL4 and IL5, but with no significant changes in IFNγ, TNFα, IL10, or IL17 expression by PBMCs or CD4 + T cells, indicating that the immunomodulation afforded by mitoxantrone treatment in MS acts through the enhancement of Th2-type cytokines [130] . Currently, a head-to-head race for approval had initially developed between two under spotlight oral immunomodulatory agents-fingolimod and cladribine ( Figure 2 ) [131] . Fingolimod (FTY720/Gilenya, Novartis), an S1PR modulator [132] , is under the spotlight because it has completed phase III trials [133] and has been approved by the US Food and Drug Administration as the first oral, first-line treatment for relapsing MS [134, 135] . S1PR is mainly expressed by immune cells, neuronal cells, endothelial cells, and smooth muscle cells [136] [137] [138] [139] . The key roles of S1PR in angiogenesis, neurogenesis, and the regulation of immune cell trafficking, endothelial barrier function, and vascular tone were demonstrated with the genetic deletion of S1pr in a murine model [140] [141] [142] . The immunomodulatory effect of fingolimod acts in two pathways. In one pathway, it inhibits the function of S1PR, which facilitates the CC-chemokine receptor 7-(CCR7-) mediated retention of lymphocytes in the lymph nodes, including naïve T cells and central memory T cells, but not effective memory T cells. This significantly reduces the infiltration of inflammatory cells into the CNS [143, 144] and reduces the numbers of autoreactive Th17 cells that are recirculating via the lymph and blood to the CNS [145] [146] [147] . The second pathway prohibits neuroinflammation via the modulation of the . This activation results in the increased production of proinflammatory cytokines, which lead to the aberrant activation of Th1 and Th17 proinflammatory responses. Activated encephalitogenic adaptive immune effectors (such as Th1 cells, Th17 cells, CD8 + cells, and B cells) express surface molecules that allow them to penetrate the blood-brain barrier and to enter the central nervous system (CNS). The presence of autoreactive immune effectors, together with abnormally activated CNS astrocytes and microglia, leads to the increased production of reactive oxygen species, excitotoxicity, autoantibody production, and direct cytotoxicity, which are all involved in the demyelination and axonal and neuronal damage that is present in patients with MS. Potential therapeutic interventions at different levels of the immunopathological cascade are shown in the filled yellow boxes (cytotoxic T lymphocytes [ [55] .). S1PR1 expressed on oligodendrocytes, neurons, astrocytes, and microglia [76, 148, 149] . Another oral immunomodulatory drug Cladribine (2-chlorodeoxyadenosine) is a synthetic chlorinated deoxyadenosine analog [150] that is activated by intracellular phosphorylation in specific cell types, resulting in preferential and sustained reduction of peripheral T and B lymphocytes, mimicking the immunedeficient status of hereditary adenosine deaminase deficiency [151] . Orally administered cladribine shows significantly efficacy in patients with RR-MS [152] . Relative to placebo, oral cladribine reduces relapses by 55-58% and has an impact on disability progression and all MRI outcome markers in patients with RR-MS [152] [153] [154] . Nevertheless, to exactly weight the benefits of both novel immunomodultory agents against the potential risks is necessary and must be monitored continually. These advances in identifying unique therapeutic targets for MS have instigated numerous phase II and phase III clinical trials, for example, trials of various mAbs, including those directed against CD52 (alemtuzumab), CD25 (daclizumab), and CD20 (rituximab), and trials of disease-modifying therapies, such as teriflunomide, laquinimod, and fumarate [135, 155] . For example, alemtuzumab, a humanized mAb, targets the surface molecule CD52 on all T-cell populations and other cellular components of the immune system, such as thymocytes, B cells, and monocytes [156] . Offner reported that estrogen and its derivatives exert neuroimmunoprotective effects against EAE and that E2 upregulates the expression of Foxp3 and Ctla4, which contribute to the activity of Tregs, suggesting the therapeutic application of estrogen to MS [157] . Papenfuss et al. also demonstrated that estriol (E3), a pregnancy-specific estrogen, has therapeutic efficacy in MS and EAE and they confirmed that E3 protects mice against EAE by inducing DCs to increase their expression of inhibitory costimulatory markers (PD-L1, PD-L2, B7-H3) and deviate towards a Th2 phenotype [158] . Peroxisome proliferator-activated receptors (PPARs) are members of the nuclear hormone receptor superfamily, which includes receptors for steroids, retinoids, and thyroid hormones, all of which are involved in the immune response [159] . Natarajan et al. demonstrated that PPARγ agonists inhibit EAE by blocking IL12 production, IL12 signaling, and Th1 cell differentiation [160] . Kanakasabai et al. further demonstrated that the PPARδ agonists ameliorate EAE by blocking IFNγ and IL17 production by Th1 and Th17 cells. The inhibition of EAE by PPARδ agonists is also associated with reductions in IL12 and IL23 and increases in IL4 and IL10 expression in the CNS and lymphoid organs. This indicates that PPARδ agonists modulate the Th1 and Th17 responses in EAE, and suggests their use in the treatment of MS and other autoimmune diseases [161] . Minocycline, an oral semisynthetic tetracycline antibiotic, can penetrate the CNS and has interesting pleiotropic biological functions and neuroprotective effects, including in demyelinating diseases such as MS [55] . Nikodemova et al. have shown that minocycline attenuates EAE in rats by reducing T-cell infiltration into the spinal cord and downregulating LFA-1 on T cells, but without modifying the production of dominant cytokines [162] . Zabad et al. demonstrated in a cohort study the impact of oral minocycline on clinical and MRI outcomes and serum immune molecules during the 24 months of open-label minocycline treatment. No relapses occurred between months 6 and 24, and the levels of the p40 subunit of IL12 were elevated during the 18 months of treatment, which might have counteracted the proinflammatory effects of IL12R. The downregulation of MMP9 activity was reduced by minocycline treatment [163] . Brines et al. have demonstrated that EPO mediates neuroprotection against experimental ischemic brain injury [164] . Agnello et al. have shown that EPO exerts an antiinflammatory effect that ameliorates EAE [165] . Yuan et al. also demonstrated that EPO retains its immunomodulatory capacity in both the periphery and the inflamed spinal cord by promoting a massive expansion of Treg cells, inhibiting Th17 polarization and abrogating the proliferation of antigen-presenting DCs [166] . We observed significantly reduced levels of both Th1 and Th17 cells in the CNS and a significantly increased proportion of splenic Tregs in EPOtreated Mog-EAE mice. We also demonstrated that MOGspecific T-cell proliferation was suppressed in the EPOtreated group [167] . The immunomodulatory mechanisms of immunomediated therapeutic agents are not fully understood. Here, we report our current understanding of the immunomodulatory effects of clinically proven and clinically tried agents, and of potential candidate agents, such as decoy receptor 3 (DcR3). We have selectively reviewed their immunomodulation in EAE and MS. Demjen et al. showed that the neutralization of CD95L (FasL) promoted axonal regeneration and functional improvement in an injured animal model, suggesting that this therapeutic strategy may constitute a potent future treatment for human spinal injury [168] . DcR3 is a recognized member of the TNFR superfamily and is predominantly expressed in tumor cells, allowing them to evade immune attack [169] . DcR3 is a soluble receptor that binds to members of the TNF family and can competitively inhibit the binding of TNF to TNFRs [170] . FasL, LIGHT, and TNF-like molecule 1A (TL1A) are all confirmed ligands of DcR3 [171, 172] . When DcR3 binds to FasL, it inhibits FasL-induced apoptosis [169] . It has also recently been shown that DcR3 counteracts the effects of Th17 cells by interfering with FasL-Fas interactions [173] . We have demonstrated that DcR3 ameliorates EAE by directly counteracting inflammation and downregulating Th17 cells in situ [174] , implying that DcR3 downregulates the Th17 response and inhibits the inflammation of the CNS in situ during EAE by blocking ligand-receptor interactions, such as Fas-FasL, DR2-LIGHT, and/or DR3-TL1A. Therefore, we introduce DcR3, another immunomodulatory molecule, as a potential candidate for consideration in the clinical treatment of MS. In summary (Figure 2 ), these immunomodulatory agents and neuroprotective therapies for MS have great value as clinical agents, to be tested in clinical trials or preclinical studies, and in the development of novel therapeutic strategies for MS [55] . MS is the most common disabling CNS disease in young adults. It is characterized by recurrent relapses and/or progression, which are attributable to multifocal inflammation, demyelination, and axonal pathology within the brain and/or spinal cord [175] . The effector Th cells play a well-recognized role in the initiation of autoimmune tissue inflammation, and these autoreactive effector CD4 + T cells have an established association with the pathogenesis of this disorder [17] . However, in models thought to be driven by Th1 cells, mice lacking the hallmark Th1 cytokine IFNγ were not protected from EAE but tended to display enhanced susceptibility to this disease [26] . The identification of Th17 cells has shed light on this apparent discrepancy. Like Th1 cells, polarized Th17 cells have the capacity to cause inflammation and autoimmune disease. A deficiency of the Th17-related cytokine IL23, but not of the Th1-related cytokine IL12, induces resistance to EAE, implying that Th17 cells are the chief contributors to EAE/MS [28] , whereas Th1 cells can consistently transfer EAE disease [16, 17] . Komiyama et al. demonstrated that EAE was significantly suppressed in Il17 −/− mice, manifested as delayed onset, reduced maximum severity, ameliorated histological changes, and early recovery [176] . However, the outcomes have varied when the differentiation and/or functions of Th17 cells have been blocked in clinical trials of human autoimmune diseases, with notable success only in psoriasis and Crohn's disease, but negative results in relapsing/remitting MS. The strategy of inhibiting the Th17 response has had even less support in preclinical studies in animal models [177] . These data raise the questions of whether MS is mediated solely by Th1 cells or solely by Th17 cells, whether it is mediated by both pathways, or whether perhaps it is mediated by neither pathway [175] . There is growing evidence that autoreactive T cells (particularly Th1 and Th17 cells) participate in the pathophysiology of MS. Although the exact roles of Th1 and Th17 cells in the development of MS lesions are not well understood, it appears that both these effector T-cell populations can cause CNS inflammation and demyelinating lesions in MS and EAE [50, 178] . Our increasing understanding of the immunopathogenic roles of Th1, Th2, and Th17 cells and Tregs in MS/EAE should facilitate the development of novel immunomodulatory therapeutic approaches to MS [179, 180] . The treatment of MS has always been hampered by the untoward adverse effects caused by immunosuppression with agents such as natalizumab [128] . Currently approved disease-modifying treatments achieve their effects primarily by blocking the proinflammatory response in a nonspecific manner. Their limited clinical efficacy calls for a more differentiated and specific therapeutic approach. We can confidently say that IFNβ, GA, and mitoxantrone are fairly clinically effective for MS patients. The addition of estrogen(s) or minocycline has also shown benefits in the treatment of MS. We have established the protective effects of DcR3 and EPO against EAE [174, 181] , but further evidence is required before they can be used clinically for the treatment of MS. More immunomodulatory therapeutic agents are currently in clinical trials, including fingolimod (FTY720), alemtuzumab, and rituximab add-on therapies [182] . The extensive clinical application of these potential novel immunomodulatory therapeutic agents will be under close scrutiny in the near future.

Towards cross-lingual alerting for bursty epidemic events

BACKGROUND: Online news reports are increasingly becoming a source for event-based early warning systems that detect natural disasters. Harnessing the massive volume of information available from multilingual newswire presents as many challanges as opportunities due to the patterns of reporting complex spatio-temporal events. RESULTS: In this article we study the problem of utilising correlated event reports across languages. We track the evolution of 16 disease outbreaks using 5 temporal aberration detection algorithms on text-mined events classified according to disease and outbreak country. Using ProMED reports as a silver standard, comparative analysis of news data for 13 languages over a 129 day trial period showed improved sensitivity, F1 and timeliness across most models using cross-lingual events. We report a detailed case study analysis for Cholera in Angola 2010 which highlights the challenges faced in correlating news events with the silver standard. CONCLUSIONS: The results show that automated health surveillance using multilingual text mining has the potential to turn low value news into high value alerts if informed choices are used to govern the selection of models and data sources. An implementation of the C2 alerting algorithm using multilingual news is available at the BioCaster portal http://born.nii.ac.jp/?page=globalroundup.

As electronic data expands, online reports are coming to represent a new modality in early warning surveillance for natural disasters such as epidemics [1] , typhoons and earthquakes [2, 3] . Recent studies in disease surveillance such as [4] have shown that significant challenges still exist for fine-grained automated understanding of event dynamics. Since 2006, BioCaster [5] has been performing gathering, semantic analysis and mapping of global news reports to provide a near-real time summary of human epidemics. The system is used regularly by both national and international health agencies as well as a growing base of individual users. Recent advances include expanding the number of diseases to include animal and crop pathogens as well as extending the number of languages from 4 to 13. With the increase in data came an understanding that public health analysts needed more help finding novel trends in the event stream. In order to support the task of detecting the unusual, we compare five widely used temporal aberration detection algorithms to look for spikes in the news event stream. This paper builds on our previously reported study for monolingual news alerting [4] by seeking to explore the hypothesis that cross-lingual events from text mining can provide improved detection rates. Although we focus here on newswire as a source we believe the results should have applicability for other unverified reports such as email lists and the rapidly developing space of user generated content. The 2009 H1N1 pandemic illustrated how dependent each country is on the surveillance capacity in other states. Reducing public health risk depends on an overall strengthening of global health event monitoring as well as locally available sources such as clinical data and over-the-counter sales data. The Web provides a low cost surveillance infrastructure that has been shown to offer a timely means of detecting epidemics such as SARS [6] that is often several days ahead of the official reporting curve. In addition to work on BioCaster, there is a small but growing body of work looking at the issues of online public health monitoring such as GPHIN [6] and MedISys/PULS [7] . However, studies providing details of recall/precision/timeliness for end user tasks in media-based health surveillance are still surprisingly limited. To the best of our knowledge no previous study has explored the multilingual effects in this area. Several characteristics of early epidemic detection make the problem particularly challenging. Firstly, we want to catch epidemics as early as possible before they develop into humanitarian crises; Secondly, not every epidemic is of equal importance -those that are of most concern to the international community are described by the International Health Regulations [8] ; Thirdly, patterns of media coverage are complex [9] , at times focussing on dramatic and emotive imagery, at others prioritizing the reader's security and economic interests. In many ways the connection between media interest and the population at risk is often blurred. How is this work different to various research in topic detection and tracking (TDT) [10] that has been undertaken for the last 14 years? Whilst both tasks look for events that are highly localized in time and space, the task we undertake begins with a predefined event semantics and a desire to distinguish the unexpected from the typical. Put another way, bursts in media interest do not always correspond to public health significance. The stream of work here seeks to uncover underlying trends and factors. Neither is this task entirely the same as TDT's first topic detection since we measure performance partly by the number of days before the silver standard that we can capture an event. In general it is extremely difficult to determine ground truth for the actual numbers and durations of disease outbreaks. As a silver standard we have chosen the best publicly available human network of reporters which is ProMED-mail [11] . ProMED-mail is a program of the International Society for Infectious Diseases with many expert volunteer reporters globally and a sophisticated staged editorial process. Outbreak reports are distributed to 40,000 subscribers by email, RSS feed and Web portal -precisely the audience we target in our automated system. In this study we have used quite coarse-grained granularity by choosing countries and days as the units. This is due to the current limits of reliable location detection in the system and also the frequency of news that we observe. The recorded time for each event was normalized to system download time which takes place every hour of each day. Evaluation uses the standard classification test measures of sensitivity (recall), specificity, positive predictive value (PPV or precision), negative predictive value (NPV) and timeliness. We also measured the average number of system alarms per 100 days and compared this to the silver standard. The F-measure (F1) is calculated in the usual way as the harmonic mean of sensitivity and PPV. As in our previous study, the standard for a true positive was to obtain a system alert on a country-disease event on or before the silver standard alert. To allow for compatibility and comparison we kept the period for a qualifying system alert as up to 7 days prior to and including a qualifying ProMED report on the same topic. Other history period lengths might be more or less effective but were not the target of the investigation in this study. True positives were increased by 1 if there was any system alert that fell within the 7 day period. Multiple system alerts did not count twice. False positives were increased by 1 for each system alert that fell outside of the 7 day window. False negatives were counted as the number of qualifying alert periods when there were no system alerts. True negatives were counted as the number of days outside of any qualifying alert period when no system alert was given. In testing we tried to maximize F1 together with timeliness. Figure 1 shows the 16 event streams that we explored. The events chosen for this study were determined based on diversity of geographical and media coverage rather than random selection. The 16 event streams contain 2064 surveillance days with 153 events (7.4% of alerting days) (Note that system data from the study will be made publicly available online for re-use via the GENI database interface on BioCaster). Since we wanted to explore the hypothesis that linguistic coverage in multiple languages could strengthen detection rates and timeliness we compared English news coverage against all languages including English for each of 16 disease outbreaks. English was chosen as the baseline because of its overall geographic representativeness. An alternative and perhaps more realistic approach might have been to use the native language for each outbreak country as the baseline which we will consider in future investigations. Because cross-lingual events on the 13 languages were only available in our system from December 2009, the trial period was from January to May 2010. ProMED reports used in the silver standard excluded those that fell outside our case definition, based on the International Health Regulations [8] decision tree instrument. For example, requests for information, reports primarily focussed on control measures and aggregated summary reports not arising from specific events. The text mining system we explored involves a semantic pipeline of modules running on a high throughput cluster computer with 48 Xeon cores. Throughput is approximately 9000 articles per day. System news was gathered from multiple news sources through Google News and MeltWater News as well as specialized sources such as the European Media Monitor, IRIN and ReliefWeb. (Note that no ProMEDmail messages were included in the system data for this study using a block on the Internet domain and message title). In total this gives us access to over 80,000 news sources globally. The languages used in the study (in ISO-639-1) are: ar,zh,nl,en,fr,de,it, ko,pt,ru,es,vi and th. Underlying the system is a publicly available multilingual application ontology [12] which is used within the rule books to make basic inferences such as countries from names of provinces, or diseases from causal pathogens. The BioCaster ontology (BCO) rules also allow us to unify variant forms of terms such as the 11 forms of A(H1N1). After data sourcing, translation takes place from the twelve non-English languages used in this study using Google's online translation system. As a quality reference point we refer to a recent large-scale evaluation of machine translation for European language pairs [13] . In this study on news texts it was found that across a wide variety of metrics Google's online system consistently performed among the highest quality systems for Spanish-English, French-English and German-English language pairs. Following machine translation, text classification using Naive Bayes (F1 0.93) removes non-disease outbreak news before text mining is applied. Rules are based on a regular expression matching toolkit called the Simple Rule Language [14] and divided between 18 entity types and template rules. The final structured event frames in XML includes slot values normalized to BCO root terms for disease, pathogen (virus or bacterium), time period, country and province. Additionally we also identify 15 aspects of public health events critical to risk assessment. For the purpose of this study we only made use of disease and country slots. Events in the 13 languages are treated in this study as being part of a univariate model for comparison purposes against English events. Latitude and longitude of events down to the province level are found automatically using Google's API up to a limit of 15000 lookups per day, and then using lookup on 5000 country and province names harvested from Wikipedia. We experimented with a range of popular models for early alerting used in the public health community: the Early Aberration and Reporting System (EARS) quality control chart models C3, C2 and W2 as well as the F-statistic and the Exponential Weighted Moving Average (EWMA). All were implemented in Excel for the purpose of this study. The models are what might be termed 'snapshot' models because they all use short 7 day baselines that assume a relatively stationary background, i.e. ignoring medium to long term periodic variations such as seasonal cycles. The baselines are used to predict future trends against which the current day values are compared. All models also use a 2 day 'guard period' just before the target day t to prevent the current day's data from being included in the baseline. All models use a minimally supervised method by setting a threshold parameter which we determined using the same 5 held out data sets used by [4] . These were 0.2 (C2 and W2), 0.3 (C3), 0.6 (F-statistic) and 2.0 (EWMA). A minimum standard deviation was set at 0.2 and a frequency purge was applied to remove singleton events, i.e. those with counts of 1 per day. The EARS algorithms [15] are based on cumulative sum calculations commonly used in quality control. C2 triggers an alert when a test statistic S t exceeds a number k of standard deviations above the baseline mean: where C t is the event count on the target day, µ t and s t are the mean and standard deviation of the counts during the baseline period. We set k to 1 for all experiments. C3 is a modified version of C2 so that the previous 2 observations (within the guard period) are added to the test statistic if the counts on those days does not exceed a threshold of 3 standard deviations plus the mean on those days. The rationale here is to extend the sensitivity of C2. W2 [16] is a stratified version of C2 which compensates for weekend data outages by removing Saturday and Sunday data counts from the baseline. Alerting though can take place on any day. The calculation for the F-statistic [17] is: Collier Journal of Biomedical Semantics 2011, 2(Suppl 5):S10 http://www.jbiomedsem.com/content/2/S5/S10 where s t 2 approximates the variance during the testing window and s b 2 approximates the variance during the baseline window. Calculation is as follows: EWMA Unlike other models in our test, the EWMA provides for a non-uniformly weighted baseline by down-weighting counts that are on days further from the target day: where 1 >l > 0 is a parameter that controls the degree of smoothing. The optimal level found from held out data was found to be 0.2. The test statistic is calculated as: . m s l l 2 0 5 As above, µ t and s t are the mean and standard deviation on the baseline window. Interestingly we found that approximately 80% of news reports covered only about half the ProMED-mail alert disease-country topics, implying that the remaining 20% of news has to provide coverage for almost half the topics. Surprisingly, the trend was broadly similar for both English and all language news. Although the sample size is relatively small, given that the events we chose were from all regions of the world, this implies that having news in more languages may have a deepening effect rather than a broadening effect on event coverage. The three notable exceptions were in the cases of FMD in China (e4 in Figure 1 ), Dengue in Brazil (e12) and Dengue in Bolivia (e13). Results for global events on English (Table 1) show an advantage for the F-statistic if we are primarily concerned with sensitivity (recall) and alerting rates (shown in column B ). However the F-statistic has a clear disadvantage with PPV (precision) which impacts heavily on the number of false alarms. This can be seen clearly by comparing the alarm rate per 100 days of 16.2 in column A with the ProMED average of 7.4. Both advantages and disadvantages are amplified when we add cross-lingual events. Whilst the F-statistic has the highest overall F1, its high rate of false alarms reflected in the PPV makes it potentially an undesirable choice. If we seek for the best balance of F1 and timeliness with a minimum of false alarms then C3 looks like a more desirable alternative. Cross-lingual event capture seemed to extend sensitivity in all models, improving F1 and timeliness. To see if we could harden our intuitions about these effects we looked specifically at South East Asia -a region where we would expect the representation of Chinese to be proportionately greater than English. Table 2 shows results which largely mirror those for the world as a whole. The noticable exception though is that EWMA shows a large drop in performance. Although the sample size is limited, the data suggests trends in model performance. C3 seems to perform best when we consider that the high false alarm rate for the F-statistic could desensitize users. Cross-language events generally seem to improve F1 performance by several points across most models except for EWMA. The benefits come from an extension in sensitivity but could be focussed on topics where we already have large coverage of English news. This is not to say that multilingual news is not useful, as we comment below, it could be that it has a greater role to play in extending detection rates of novel events at lower levels of geographic granularity than the country. Beyond the cross-lingual effects, drill down analysis revealed that bag-of-words topic classification and event extraction using intra-sentential regular expressions were still letting through a proportion of non-events. We sampled 274 English news articles by hand from the BioCaster portal's implementation of the C2 algorithm using a 7 day baseline window and found that approximately 30% of positively classified news articles fell just below the borderline of our case definition. Commonly misclassified topics included: vaccination campaigns, factual advice on avoidance and treatment of infectious diseases, improvements in surveillance facilities and surveillance exercises. Often the articles mentioned an infectious disease and cited facts about case numbers such as "90% of cholera cases reported annually…" or an analysis of historical events. This points to a need to strengthen discourse analysis, such as inter-sentential causality and inclusion relations between events which the current intra-sentential template driven approach does not handle well. The study also raises several questions about factors in the imbalance of reporting: why did Dengue in Brazil (e12) or FMD in China (e4) receive such massive local coverage but disproportionately less in the English media? Why did cholera in Angola (e9) or influenza in Romania (e8) receive comparatively low coverage overall? We also observed that the USA epidemics (e15 and e16) were widely reported in English but not so greatly in other languages. In order to illustrate the potential complexity of the task we provide a detailed drilldown analysis of one of the outbreaks in our data set, i.e. cholera in Angola. Just to put the reporting of this outbreak into context: Angola itself is a former Portuguese colony which has suffered major outbreaks (e.g. 2006 to 2008) of cholera due to poor sanitation, drinking water infrastructure and environmental conditions. Although UNI-CEF has commented on recent advances, the country remains at risk, especially during the rainy season from January to mid-May. In this case BioCaster was more successful for English than for the multilingual system because a false spike of reports occluded subsequent true positives. In the case of the silver standard report on 19/3/2010, the cited English source was not detected but its Spanish translation was found a few days later -still much earlier than the ProMED-mail report. The example is a relatively special case that illustrates an event that was not widely re-reported. The reports were made in English, Portuguese, French and Spanish from Angop. Externally, the 4/3/2010 article from Angop was republished in http://allafrica. com and http://africanseer.com on the 4th March. It was also referenced in a blog by the Namibia online community. Automated health surveillance using text mining is not intended as a substitute for skilled human analysts but as these results show, it does have the potential to reduce their information burden if informed choices are used to govern the selection of models. In order to help guide users in the significance of news events we implemented the C2 algorithm for multilingual news alerting in the 'Global Roundup' section of the Bio-Caster Web site at http://biocaster.nii.ac.jp. The results, updated each hour, show the test statistic value, the disease, country, province, focus species daily news frequency and the baseline mean and standard deviation. Additionally, citation links are provided to the news articles with a list up of all languages that contributed to the alert. The output is available as both RSS and a Twitter feed. Obvious improvements to the techniques described here could take place by modeling lower geographic granularity and reducing size differences between geo-units. More sophisticated approaches might incorporate proximity information between events or model how events propagate through news space. A more subtle effect of the granularity restriction is that the models we presented do not allow us to follow what might be called 'late warning' signals. i.e. follow on events within the country's borders. For this reason detecting events below the country level is desirable. Future work will need to concentrate on maximizing system sensitivity to overcome the fragmentation of the event distribution that occurs when we bucket events into smaller geographic units.

Reducing mortality in sepsis: new directions

Considerable progress has been made in the past few years in the development of therapeutic interventions that can reduce mortality in sepsis. However, encouraging physicians to put the results of new studies into practice is not always simple. A roundtable was thus convened to provide guidance for clinicians on the integration and implementation of new interventions into the intensive care unit (ICU). Five topics were selected that have been shown in randomized, controlled trials to reduce mortality: limiting the tidal volume in acute lung injury or acute respiratory distress syndrome, early goal-directed therapy, use of drotrecogin alfa (activated), use of moderate doses of steroids, and tight control of blood sugar. One of the principal investigators for each study was invited to participate in the roundtable. The discussions and questions that followed the presentation of data by each panel member enabled a consensus recommendation to be derived regarding when each intervention should be used. Each new intervention has a place in the management of patients with sepsis. Furthermore, and importantly, the therapies are not mutually exclusive; many patients will need a combination of several approaches – an 'ICU package'. The present article provides guidelines from experts in the field on optimal patient selection and timing for each intervention, and provides advice on how to integrate new therapies into ICU practice, including protocol development, so that mortality rates from this disease process can be reduced.

Sepsis is the tenth most common cause of death in the US [1] . A recent US study reported that severe sepsis accounts for in excess of 215,000 deaths annually from a total population of approximately 750,000 patients-a mortality rate of approximately 29% (with published studies quoting a range of 28-50%) [2] . This persistent, high mortality rate is clearly unacceptable, given that it ranks sepsis above some of the higher profile causes of in-hospital death, including stroke (12-19% risk of death in the first 30 days) and acute myocardial infarction (AMI) (8% risk of death in the first 30 days) [3] . Moreover, the actual number of deaths associated with the condition may be even higher than current estimates suggest. Many sepsis patients have at least one comorbidity and deaths are often attributed to these conditions rather than to sepsis [4] [5] [6] . Unfamiliarity with the signs and symptoms of sepsis may further hinder accurate diagnosis. There are many possible reasons for this high mortality. Sepsis is certainly a complex disease state; the pathophysiology is only now beginning to be unraveled, and it is complicated by heterogeneous presentation (possible signs of sepsis are presented in Table 1 ). While none of these signs alone is specific for sepsis, the otherwise unexplained presence of these signs should signal the possibility of a septic response. Many cases of sepsis are recognized late, and patients are often inappropriately treated before entering the intensive care unit (ICU) by physicians unfamiliar with the signs and symptoms of the condition. Furthermore, treatment may be initiated by any of a number of physicians (anesthetists, hematologists, intensivists, infectious disease specialists, pulmonologists, and emergency physicians). There are presently various defined supportive strategies for treating patients with sepsis, but improvements are needed to reduce the unacceptably high mortality rate. Moreover, as with other areas of medicine, the application and integration of new but proven strategies for reducing morbidity and mortality into clinical practice has been slow. Encouraging new data have recently been presented on new approaches to the management of patients with sepsis. Many of these approaches attempt to modulate or interrupt the sepsis cascade and to address the cause of multiorgan dysfunction. Although many of these approaches are in early phases of development (e.g. antibodies to tumor necrosis factor [TNF] alpha, bactericidal permeability increasing protein, high-flow hemofiltration to remove circulating inflam-matory mediators, platelet-activating factor acetyl hydrolase, and antielastases), other approaches are more advanced and are already beginning to impact on outcomes in the ICU. At a roundtable discussion in London in June 2002, Professor Jean-Louis Vincent brought together five experts to discuss more effective implementation of five exciting new interventions in the ICU setting to decrease the unacceptable burden of mortality in patients with severe sepsis. Each of the roundtable panelists is a highly respected physician in the world of sepsis and critical care medicine. The interventions discussed encompassed low tidal volume in patients with acute lung injury (ALI)/acute respiratory distress syndrome (ARDS) (Edward Abraham), early goal-directed therapy (EGDT) (Emanuel Rivers), drotrecogin alfa (activated) (Gordon Bernard), moderate-dose corticosteroids (Djillali Annane), and tight control of blood sugar (Greet Van den Berghe). The purpose of the roundtable discussion was to provide guidance for clinicians on the integration of new interventions into the ICU to reduce the mortality in sepsis, on appropriate patient selection for these interventions, and on appropriate timing of these interventions. The present review reports the discussions and recommendations of the panel. The overall 30-day mortality in the ICU is typically ~20% [8] . The 30-day mortality in the population with severe sepsis, defined as sepsis with organ dysfunction, is 30-50%. It is clear from this figure that severe sepsis contributes disproportionately to the overall 30-day mortality in the ICU and compares unfavorably with some of the higher profile acute killers in hospital (e.g. stroke and AMI) [3] . Despite the general improvements in medicine overall, this mortality rate has remained essentially unchanged for the past 25 years. This has contributed to a feeling of pessimism among Table 1 Possible signs of sepsis (adapted from [ intensivists and other medical professionals regarding treatment prospects for severe sepsis, and a reluctance to rapidly incorporate new interventions into clinical practice [9] . Although the sepsis mortality rates are unacceptable, they camouflage some significant developments that are and have been occurring for hospital patients, for the general ICU population and, particularly, for those with severe sepsis. Direct comparison of mortality rates among patients with identical Acute Pysiology and Chronic Health Evaluation (APACHE) scores in the placebo arm of anti-TNF or anti-endotoxin studies published 10-15 years ago [10] [11] [12] with more recent studies [13, 14] , demonstrates that the mortality rate is much lower in more recent studies. Interestingly, this decrease was apparent even before the five interventions discussed in the present article were published, reflecting improvements in the general supportive care of sepsis patients. Indeed, the panel contends that mortality from septic shock has already been reduced. Some patients who in the recent past would have died from severe sepsis or septic shock do not reach the ICU now because they are well managed on the wards, in the emergency department, and even during preoperative and postoperative care. For example, those sepsis patients that receive prompt antibiotic therapy have a 10-15% lower mortality rate than those who receive antibiotic therapy later in their care [15] . Progress is also being made in diagnosing sepsis: more patients are being tested to identify the source of infection and the pathogens involved, supportive care measures have been improved (e.g. hemodynamic support), and other measures have been put in place to reduce the incidence of nosocomial infections (e.g. reducing the need for pulmonary artery catheters by using echo techniques to assess cardiac function). There has also been a realization of the importance of specially trained intensive care physicians in the ICU. It has been internationally recognized that changing the ICU from an 'open format', whereby patients are cared for by their admitting physician, to a 'closed format', whereby patients are managed by appointed intensivists, reduces mortality rates [16] . Although the mortality rate is beginning to decline, it still remains unacceptably high. Furthermore, the number of patients with severe sepsis and septic shock is increasing; people are living longer, and there has been a rise in the number of immunocompromised patients due to aggressive cancer therapy and the increased prevalence of HIV. In-hospital AMI-associated mortality rates averaged approximately 25-30% in the 1960s [3] . This clearly unacceptable mortality rate was addressed by the development of a number of new pharmacological and mechanical interventions together with improvements in supportive care. In the landmark Second International Study of Infarct Survival trial, published in 1988, 17,187 suspected AMI patients were treated with either streptokinase or aspirin, with both drugs, or with neither. The mortality rate in the combination group of this trial was 8%, compared with 13.2% in those patients given neither streptokinase nor aspirin [17] . Cardiologists have effectively implemented multiple pharmacologic and supportive care interventions to reduce mortality in AMI from 25-30% to 8% and lower. Not satisfied with this already remarkable figure, they are trying to reduce it further. Physicians treating patients with sepsis are clearly faced with a very different situation to those treating patients with AMI, and so direct comparisons are not possible. However, several factors have contributed to the success of AMI therapy and possibly to the lack of such success in sepsis (Table 2) . Sepsis is undoubtedly complicated. However, many of the lessons that have been learned through effective application of therapies in other disease states can be applied to severe sepsis. Furthermore, the encouraging data that are beginning to appear in the literature indicate that sepsis may not be as intractable to treat as once thought. The following sections provide salient information on five interventions that have shown a significant positive impact on mortality rates in sepsis, severe sepsis, septic shock, or sepsis-related diseases in recent clinical trials. The interventions were presented at the roundtable by one of the principal investigators of the key trial of the intervention. Each section concludes with recommendations for the integration of the particular intervention into clinical practice. The traditional approach in patients with ALI/ARDS is to ventilate using tidal volumes between 10 and 15 ml/kg body weight, almost twice the average tidal volume at rest (7-8 ml/kg body weight), and to maintain a low positive endexpiratory pressure (PEEP). The purpose of this approach is to achieve normal values for the pH and partial pressure of arterial carbon dioxide. However, this method leads to high inspiratory airway pressures and to excessive stretch of the aerated lung. In 1997, Tremblay et al. examined the effect of ventilation strategy on lung inflammatory mediators in the presence and absence of a pre-existing inflammatory stimulus in Sprague-Dawley rats [18] . In both stimulated and nonstimulated groups, the presence of inflammatory mediators (TNF-α, IL-1β, IL-6, IL-10, macrophage inflammatory protein 2, and IFN-γ) was highest in those rats ventilated with a large tidal volume and zero PEEP. Furthermore, in a study by Ranieri et al. in 1999 [19] , the concentration of inflammatory mediators 36 hours after randomization of the groups was significantly lower in the lung-protective strategy group (tidal volume, 7.6 ± 1.1 ml/kg) than in the control group (tidal volume, 11.1 ± 1.3 ml/kg) (P < 0.05). Following on from the positive results in the Tremblay et al. trial [18] , a small study (53 patients) was carried out by Amato et al. in Brazil [20] . The mortality rate was 38% in patients given 'protective' ventilation (PEEP above the lower inflection point on the static pressure-volume curve, tidal volume < 6 ml/kg ideal body weight, driving pressures < 20 cmH 2 O above the PEEP value, permissive hypercapnia, and preferential use of pressurelimited ventilatory modes) compared with 71% in patients on conventional ventilation (P < 0.001). This impressive reduction in mortality was tempered by the higher than normal mortality level in the control group, prompting the National Institutes of Health-funded Acute Respiratory Distress Syndrome Network to set up a similar, larger (861 patients), prospective, multicenter, randomized trial in the US [21] . For a summary of the protocol used in this study, see Appendix 1. The trial was stopped after the fourth interim analysis because the use of lower tidal volumes was found to be associated with a significantly reduced mortality (P = 0.005 for the difference in mortality between groups). The primary endpoints were mortality prior to hospital discharge with unassisted breathing and ventilator-free days (days alive, off mechanical ventilation, between enrollment and day 28). Both of these endpoints were achieved (Figs 1-3 ). In addition, patients receiving a tidal volume of 6 ml/kg ideal body weight had increased organ failure free days and lower IL-6 levels. ALI is seen in 25-42% of patients with sepsis [22] . Although the approach has only been tested in patients with ALI/ARDS, a tidal volume of 6 ml/kg ideal body weight is at the lower end of the range of physiologic ventilation. Hence, this approach should be suitable for most patients in the ICU setting. Furthermore, as many patients with severe sepsis or septic shock progress to frank ALI/ARDS, the panel believes that low tidal volume therapy is a valid option in these patients, and an option that may indeed prevent the development of ALI/ARDS. Although patient selection in the clinical trial specified both blood gas and lung infiltrate criteria, at least 90% of patients in the general ICU setting meet the criteria for blood gas but Table 2 A comparison of acute myocardial infarction (AMI) and sepsis Market issues Significant publicity surrounding and general awareness Lack of understanding among physicians and the of the condition; large trials general public Diagnosis A relatively straightforward and relatively common Complicated by a long list of signs and symptoms diagnosis (electrocardiogram, enzymes, troponin), and and few objective tools for validation one that can be made by generalists, not just cardiology specialists Generally single organ disease (notable exception when Often chronic or acute comorbidities complicated by cardiogenic shock) Generalists have been taught to recognize the signs Sepsis patients often come 'second hand' from a and symptoms of AMI; initial treatment is usually specialist who may not be appropriately trained to provided by emergency physicians, who are trained diagnose, manage, and refer patients with sepsis to treat these patients Mortality prior to hospital discharge in patients receiving a tidal volume of 6 and 12 ml/kg ideal body weight. Acidosis is more likely to develop in patients with severe lung problems rather than in those exhibiting milder disease when tidal volumes are kept low. However, acidosis is seldom a clinical problem and rarely requires administration of bicarbonates. One of the issues with low tidal volume therapy is that the patients are often more uncomfortable, at least initially, when they are being ventilated with a tidal volume of 6 ml/kg ideal body weight. The patients tend to exhibit tachypnea and may become more agitated. Sedation is generally required, but the ventilator setting can be maintained. Of more concern is that ICU staff may consider a respiration rate of 40/min to be a sign of something more serious and may attempt to terminate the intervention. Education of staff is clearly essential. The strategy assessed in this trial not only includes ventilation with a low tidal volume, but also the provision of extrinsic PEEP. There may be some concern that an increased respiratory rate may result in intrinsic PEEP and hemodynamic problems (e.g. decreased cardiac filling, decreased cardiac output, and diminished blood pressure). The panel believes that auto PEEP was not an issue in the Acute Respiratory Distress Syndrome Network study. In addition, in the groups with low tidal volume, at least 10% more oxygen was required to maintain the fraction of inspired oxygen (FiO 2 ), suggesting that there was very little auto PEEP occurring. When mechanical ventilation is indicated for treatment of patients with ALI/ARDS, the tidal volume should be limited tõ 6 ml/kg ideal body weight. Goal-directed therapy represents an attempt to adjust the cardiac preload, afterload, and contractility to balance systemic oxygen delivery with oxygen demand. In patients with severe sepsis and septic shock, such an approach would seem eminently reasonable as part of general supportive measures to restore and maintain adequate cellular perfusion and to prevent organ dysfunction. In the setting of the ICU, however, supranormal and normal approaches have met with little or no success [23, 24] . It is possible that, by the time these therapies are applied in the ICU, any such intervention may have been too late. Hence, the focus has shifted towards hemodynamic optimization in the early presentation of disease, such as in the emergency department. A prospective, randomized, predominantly blinded study was initiated by the Early Goal-Directed Therapy Collaborative Group to examine the results of hemodynamic interventions in the emergency department [25] . In this study, patients were randomly assigned to either 6 hours of EGDT or to standard therapy prior to admission to the ICU. Baseline characteristics (including the adequacy and duration of antibiotic therapy) in the EGDT and standard therapy groups were not significantly different. The vital signs, resuscitation endpoints, organ dysfunction scores, and coagulation-related variables were similar in these groups at baseline [25] . However, there were some important differences between the treatment groups (see Table 3 ). Available online http://ccforum.com/content/6/S3/S1 Proportion of patients alive and off the ventilator having been ventilated with a tidal volume of 6 and 12 ml/kg ideal body weight. Median number of ventilator-free days in patients receiving a tidal volume of 6 and 12 ml/kg ideal body weight. Patients randomized to EGDT received the same therapy but, in addition, were monitored for the endpoint of central venous oxygen saturation (ScvO 2 ) > 70%. EGDT patients were given more intravenous fluids (including blood transfusions) and more inotropic support (mostly dobutamine). For more information on the protocol used in this study, see Appendix 2. Key data are presented in Table 4 . The in-hospital mortality was 30.5% in the group assigned to EGDT and was 46.5% in the group assigned to standard therapy (P = 0.009), indicating that EGDT provides significant benefits in improving outcomes in patients with severe sepsis and septic shock. During the interval from 7 to 72 hours, patients assigned to EGDT exhibited a more significant improvement in mean ScvO 2 (70.4 ± 10.7% versus 65.3 ± 11.4%), in lactate concentration (3.0 ± 4.4 mmol/l versus 3.9 ± 4.4 mmol/), in base deficit (2.0 ± 6.6 mmol/l versus 5.1 ± 6.7 mmol/l), and in pH (7.40 ± 0.12 versus 7.36 ± 0.12) than patients assigned to standard therapy (P ≤ 0.02 for all comparisons). During the same period, the mean APACHE II scores were significantly lower, indicating less severe organ dysfunction, in the patients assigned to EGDT than in those patients assigned to standard therapy (13.0 ± 6.3 versus 15.9 ± 6.4, P < 0.001). The protocol was based predominantly on guidelines published in 1999 by the Society of Critical Care Medicine [26] . However, these guidelines have not been universally followed in clinical practice since their publication. An increasing number of critically ill patients are presenting to, and being treated in, emergency departments [27, 28] . This is present-ing significant resource challenges in the emergency department environment. The inability to institute EGDT may thus not be a conscious decision by the clinician not to follow the Society of Critical Care Medicine guidelines. Emergency medicine in general may have to develop and formulate the cost-benefit analysis to support or implement such care in this environment in order to improve outcomes. There are sufficient evidence-based data to recommend that all patients with severe sepsis or septic shock should receive early and aggressive resuscitation based on this EGDT protocol (see Appendix 2) . It is important that the interventions are individualized to each patient. A negative or positive value indicates how the control group therapy compares with the treatment group. a P < 0.001, b P = 0.01, c P = 0.02, d P = 0.03, e P = 0.04. EGDT, early goal-directed therapy. it is possible to identify patients with profound global myocardial dysfunction who are hence at risk of impaired perfusion. These patients, almost 15% of those in the EGDT group, received dobutamine during the first 6 hours because myocardial suppression was diagnosed. Once myocardial dysfunction is corrected (and compliance improved), these patients become more suitable for volume loading, so this group received almost 3.5 liters more fluids in the first 6 hours than the control patients. Therefore, although vasopressor use was similar in the first 6 hours, patients in the EGDT group were more aggressively weaned off these agents during this period, resulting in fewer patients in this group entering the ICU on vasopressors than in the control group. The lack of aggressive volume loading in the control group led to greater use of vasopressors in patients over the subsequent 72 hours. In spite of more volume loading, the EGDT group received less mechanical ventilation over the subsequent 72 hours than in the standard treatment group. Why was cardiovascular collapse a significant cause of death in the control group? Cryptic shock (shock with normal vital signs) is a frequent occurrence in early severe sepsis and septic shock. Despite resuscitation to the goals for mean arterial blood pressure and CVP, almost 40% of control patients continued to exhibit global tissue hypoxia (decreased ScvO 2 and increased lactate levels); in these patients, there was a twofold increase in hemodynamic deterioration, requiring more mechanical ventilation, pulmonary artery catheterization, and vasopressor use in the subsequent 72 hours. How do severe sepsis and septic shock differ hemodynamically in the early stages compared with that classically described in the ICU? Patients presenting with early sepsis and septic shock are characterized by hypovolemia (low CVP), normal to increased blood pressures, and decreased cardiac output (decreased central venous oxygen saturation and low cardiac index). This is in contrast to ICU patients who are euvolemic, have high ScvO 2 , and have elevated cardiac indices [29] . What are the most important ways in which EGDT can improve outcomes? The key factors are early detection of high-risk patients in cryptic shock, early reversal of hemodynamic perturbations and global tissue hypoxia, prevention of acute cardiovascular collapse, and the possibility of preventing the inflammatory aspects of global tissue hypoxia that accompany the inflammation or infection. Severe sepsis and septic shock patients should receive early aggressive therapy to restore and maintain oxygen availability to the cells. There should also be generous use of fluids and inotropic agents titrated by appropriate hemodynamic monitoring. Background A large number of observational studies have shown that patients with sepsis have severe depletion of protein C [30, 31] . A number of studies have also shown the association of protein C depletion with high mortality in sepsis [32] [33] [34] . Furthermore, baboon studies have demonstrated that treatment with activated protein C prevents death from live Escherichia coli infusions [35, 36] . Activated protein C exerts a number of actions. Anticoagulant action includes the inactivation of coagulation factors Va and VIIIa, and the inhibition of the formation of thrombin. Profibrinolytic action allows the activity of tissue plasminogen activator (endogenous tissue plasminogen activator), by inactivating plasminogen activator inhibitor 1 and thrombin activatable fibrinolysis inhibitor. Finally, anti-inflammatory action reduces IL-6 (in vivo) and proinflammatory cytokines (in vitro). The specific mechanisms by which drotrecogin alfa (activated) exerts its effect on survival in patients with severe sepsis are not completely understood. The efficacy of drotrecogin alfa (activated) (recombinant human activated protein C) in reducing mortality in patients with severe sepsis was investigated in a large multicenter, blinded, placebo-controlled, randomized, phase III clinical trial, the Protein C Worldwide Evaluation in Severe Sepsis (PROWESS) trial [14] . All patients in the PROWESS trial received standard supportive care in addition to either drotrecogin alfa (activated) or placebo. For a summary of the protocol used in the PROWESS study, see Appendix 3. The overall mortality in patients treated with drotrecogin alfa (activated) was 24.7% compared with 30.8% in patients receiving placebo, an absolute risk reduction of 6.1% (P = 0.006) (see Fig. 4 ). The absolute risk reduction in patients with high risk of death defined by an APACHE II score ≥ 25 was 12.8% (P < 0.001). The absolute risk reduction in patients with high risk of death defined by multiple organ failure was 7.4% (P = 0.006). No substantial differences in drotrecogin alfa (activated) treatment effects were observed in subgroups defined by gender, ethnic origin, or infectious agent. Can drotrecogin alfa (activated) be used in patients on dialysis for pre-existing renal failure, a category that was specifically excluded in the PROWESS trial? No pharmacokinetic data were available on drotrecogin alfa (activated) in patients on chronic dialysis when the PROWESS trial began, so such patients were excluded from the trial. Subsequent research has shown that the pharmacokinetics of drotrecogin alfa (activated) are not substantially changed in patients on chronic dialysis. The design of the PROWESS trial allowed a maximum of 48 hours between the onset of first organ dysfunction and the receipt of drotrecogin alfa (activated) (a 24-hour window was allowed for receipt of the drug following the first confirmation of first organ dysfunction, which in turn had to have been present for no more than 24 hours). The treatment effect of drotrecogin alfa (activated) was consistent across all time intervals from meeting the entry criteria to the receipt of the study drug. Treatment with drotrecogin alfa (activated) thus does not appear to be as time critical as interventions such as tissue plasminogen activator in stroke or myocardial infarction. Because most of the experience with drotrecogin alfa (activated) was based on organ failure times less than 48 hours, treatment should not be delayed when an appropriate candidate is identified. The time window employed in the PROWESS trial should allow a full history to be taken and other tests to be performed to determine the bleeding risk. As with all anticoagulants, drotrecogin alfa (activated) is associated with a risk of severe bleeding. During the infusion period in the PROWESS trial, the bleeding rates were 2.4% in the drotrecogin alfa (activated) group versus 1.0% in the placebo group (P = 0.024). The risk of bleeding was fairly constant across most subgroups. However, severe thrombocytopenia (< 30,000/mm 3 ) was commonly associated with serious bleeding and intracerebral hemorrhage. Patients at high risk of death in the PROWESS trial were most likely to benefit from drotrecogin alfa (activated). In the PROWESS trial, the APACHE II score was the most effective predictor of risk of death and likelihood of benefit from drotrecogin alfa (activated), particularly in those patients with an APACHE II score ≥ 25. In the PROWESS trial, the number of organ dysfunctions was also an important indicator that supported an association between likelihood of benefit from drotrecogin alfa (activated) and risk of death. Two or more organ dysfunctions identify a population that responds well to therapy, and is a practical measurement. The panel believes that acute respiratory failure or hypotension unresponsive to fluid challenge should suggest the use of drotrecogin alfa (activated). However, coagulopathy, a platelet count < 80,000/mm 3 , acidosis, or low urine output alone should not suggest its use. A very large international study of 11,500 patients will be started in late 2002 to investigate the efficacy of drotrecogin alfa (activated) in patients with a single organ failure and/or APACHE scores < 25. The decision on whether to administer the drug should ultimately depend on whether the patient meets the selection criteria. A patient presenting in the emergency room with acute respiratory failure or acute cardiovascular decompensation should receive appropriate treatment there. The drawback to treatment in the emergency room is that there may not be sufficient time in which to evaluate the patient's bleeding risks. Delaying treatment for a few hours will enable more tests to be performed and a fuller history to be taken, both of which will provide a better indication of whether drotrecogin alfa (activated) is appropriate. The dose is always the same (24 µg/kg/hour), regardless of the type of organ failure or the degree of sepsis severity. In addition, the 96-hour window of treatment is always the same so that interruptions of treatment are made up at the end to maintain a total of 96 hours of treatment. Twenty-eight-day survival in patients treated with drotrecogin alfa (activated) or placebo: all-cause mortality. Do patients require any laboratory testing before they receive drotrecogin alfa (activated)? No laboratory testing was carried out in the PROWESS trial, and subgroup analysis identified no biochemical marker that conclusively indicates treatment. For example, treatment-associated reductions in mortality were observed in patients with normal protein C levels and in those with low protein C levels. Clinical criteria are recommended for the initiation of therapy. Aspirin (650 mg/day) was allowed in the PROWESS trial. Patients on glycoprotein IIb/IIIa inhibitors were excluded because no data were available regarding drug interactions and pharmacokinetics. Use of these types of agents is likely to increase the risk of bleeding with drotrecogin alfa (activated) therapy. The anticipated benefits must therefore be weighed against the potential risks. In the PROWESS trial, efforts were made to correct the international normalized ratio towards normal if it was greater than 3 at any time during infusion of drotrecogin alfa (activated). Approximately one-third of patients in the PROWESS trial received steroids at the same time as drotrecogin alfa (activated). There was no interaction with steroid use, presumably because the mechanism of action of steroids is so different from that of activated protein C. Hence, steroids should be used if they are needed, and if the patient qualifies for drotrecogin alfa (activated) the two should be used together. Drotrecogin alfa (activated) should be considered for use in all adult patients with recent onset severe sepsis or septic shock, and a high risk of death. The value of steroids in the treatment of patients with severe sepsis and septic shock has been fiercely debated for some time. Although a number of well-designed, randomized, controlled trials failed to show any benefits of steroid therapy in terms of improved survival in patients with severe sepsis (reviewed in [37, 38] ), with mortality increased in many as a result of an increased incidence of nosocomial infections, these trials were primarily investigating the efficacy of short courses of high-dose steroids. The question of whether lower doses of steroids may provide benefit in these patients has only recently been addressed. There is a relatively strong rationale for considering the use of steroids in patients with refractory septic shock. Relative adrenal insufficiency is common in patients with refractory septic shock (50-75% of patients) [39] . In addition to such relative adrenal insufficiency and the blunted response to corticotrophin, a large body of evidence indicates that sepsis and refractory septic shock are characterized by peripheral tissue resistance to corticosteroids [40, 41] . In septic patients, this can be evidenced in a variety of ways. First, global cortisol binding, which carries cortisol from the adrenal glands to the tissues, decreases in patients with severe sepsis [42] . Second, the number and binding affinity of glucocorticosteroid receptors may be reduced in patients with sepsis and severe sepsis [43] , leading to a decrease in the conversion of cortisone to its active form, cortisol, particularly by IL-2 levels in the tissues. Finally, data have been published demonstrating that moderate doses of steroids may restore cell sensitivity to vasopressors [44] . This may reduce the intensity of the inflammatory response and decrease organ dysfunction. Low-dose steroid treatment is also well tolerated [40] . This body of evidence prompted the initiation of a phase III randomized, controlled trial performed in 19 centers in France with 300 patients [45] . The aim of the trial was to determine whether moderate-dose corticosteroid therapy affected survival in patients with refractory septic shock and adrenal insufficiency. All patients had to be treated with vasopressor agents and mechanical ventilation. For a summary of the protocol used in this study, see Appendix 4. Patients were stratified according to their response to the adrenocorticotrophic hormone (ACTH) test. Nonresponders were defined by an increment in cortisol levels < 9 µg/dl or < 250 nM/l after challenge with 250 µg cosyntropin. Of the 300 patients included, there were 229 nonresponders to the corticotropin test (placebo, 115 patients; steroids, 114 patients). A significant survival benefit was demonstrated among nonresponders receiving moderate-dose corticosteroids. There were 73 deaths in the placebo group (63%) and 60 deaths in the steroid group (53%) (hazard ratio, 0.67; 95% confidence interval, 0.47-0.95; P = 0.023). No beneficial effects were observed in the subset of patients who were classified as responders. Hence, in this paradigm, the ACTH test serves as a useful prognostic measure. Since a beneficial effect was observed in the total population, however, the need for an ACTH test can be challenged and further studies are required. If an ACTH test is performed, corticosteroid administration can be started before results are received. Moderate-dose corticosteroids should be administered to patients with established refractory septic shock. What is the optimal dose for this intervention? Hydrocortisone should be given daily at a dose of 200-300 mg. Fludrocortisone should be given daily at a dose of 50 µg. What is the optimal duration for this intervention? Moderate doses of steroids should be given for 7 days. Hydrocortisone can be administered as serial boluses or as a continuous infusion. It may be that rebound phenomena at treatment discontinuation are more frequent when hydrocortisone is given as a continuous infusion. In addition, in the phase III randomized trial, hydrocortisone was given as serial boluses. The phase III randomized trial has shown that the combination of hydrocortisone and fludrocortisone increased survival. In addition, sepsis is more frequently associated with a mineralocorticoid deficiency than a glucocorticoid deficiency. Hence, fludrocortisone should be added to hydrocortisone. Administration of moderate-dose corticosteroids should be considered in cases of refractory septic shock, particularly in those with relative adrenal insufficiency. It is recommended that an ACTH test be carried out before starting the intervention. Hyperglycemia, caused by insulin resistance in the liver and muscle, is a common finding in ICU patients. It can be considered an adaptive response, providing glucose for the brain, red cells, and wound healing, and is generally only treated when blood glucose increases to > 215 mg/dl (> 12 mmol/l). Previous studies have shown that high levels of insulin-like growth factor binding protein 1 (a very good marker of lack of hepatic insulin effect) predict mortality [46, 47] . Patients with high insulin-like growth factor binding protein 1 also tend to have the lowest insulin levels, indicating that beta cell function is impaired and, therefore, not enough insulin is being produced. These results indicate that hyperglycemia may not always be adaptive and that it should be treated to avoid the onset of specific complications. Nevertheless, conventional wisdom in the ICU has been that hyperglycemia is beneficial and that hypoglycemia should be avoided. The hypothesis that hyperglycemia (> 110 mg/dl, > 6.1 mmol/l) predisposes to specific ICU complications, prolonged intensive care dependency and death was tested in a prospective, randomized, controlled trial [48] . For a summary of the protocol used in this study, see Appendix 5. Thirty-five of the 765 patients (4.6%) in the intensive insulin group died in the ICU, compared with 63 patients (8.0%) in the conventional therapy group. For further mortality data on both the length of hospital stay and the cause of death, see Tables 5 and 6 . For morbidity data, see Figure 5 . Tight control of blood sugar, as outlined in Appendix 5, requires a strict protocol for insulin administration and repeated determination of blood sugar. This is yet to be proven, and is the subject of an ongoing study. Because medical patients tend to stay in the ICU longer than surgical patients, the results from this study indicate that this intervention would be even more favorable to medical ICU patients. However, one needs to be careful with application of the algorithm in certain disease states, especially severe hepatic dysfunction and renal failure. No, all carbohydrates are included. See Appendix 5 for guidelines on feeding. The level was chosen because it is in the physiologic range for healthy people. As well as its effect on glycemia, insulin has been shown to inhibit TNF-α and macrophage inhibitory factor (when infused concomitantly with glucose). This has led to some doubts as to whether the effect in this study was due to normalization of blood glucose levels. However, multivariate analysis of all the risk factors for mortality, including severity of illness on admission, indicated that blood glucose determines the outcome; there was a 75% increase in risk of death per 50 mg/dl increase in blood glucose. It is not yet possible to determine this. Although it was blood glucose levels that were measured, the effects of insulin may in fact be on free fatty acids, as they change in parallel with S11 blood glucose. One of the key mechanisms may be prevention of hypertriglyceridemia and high concentrations of free fatty acids. It is strongly advisable to tightly control blood sugar close to physiologic levels, especially in surgical patients. Implemen-tation of this recommendation requires a well-defined ICU protocol. The interventions discussed in the present article have been applied in different patient populations and at different times in the course of the disease (see Table 7 ). It is essential for physicians to understand that these therapies are not mutually exclusive. Optimal patient management may require a combination of approaches: mechanical ventilation to preserve lung function, hemodynamic support to maintain adequate ScvO 2 , intensive insulin therapy to normalize blood sugar, steroids to provide adequate immunosuppression, and drotrecogin alfa (activated) to prevent the systemic coagulopathy characteristic of severe sepsis and, hence, to preserve organ function. A sound understanding of the indications and contraindications of these interventions will guide appropriate intervention. Similarly, the timing of therapy needs to be closely monitored. Education in the signs and symptoms of sepsis and severe sepsis should prompt early initiation of therapy. Many of the interventions discussed in this article were tested at specific Available online http://ccforum.com/content/6/S3/S1 Multiple organ failure, no sepsis focus 18 14 Multiple organ failure, with sepsis focus 33 8 Most important effects on morbidity [46] . CVVH, continuous venovenous hemofiltration; ICU, intensive care unit; NNT, number needed to treat; RRR, relative risk reduction. Despite the wealth of data to support the approaches discussed, it is clear that uptake of these interventions into clinical practice has been slow. Although there may be practical reasons for this, it would appear in many cases to involve either unfamiliarity with the data or a reluctance, or at least inertia, to change established practices (witness the necessity of proving that hypoglycemia is beneficial in ICU patients despite no good evidence to the contrary). The ICU has changed in the past 30 years; there are more tools to use and more interventions to implement. Despite application of new methods, however, outcomes have changed very little and certainly not in proportion to the changes that were expected based on the results from clinical trials. Efficient integration of new interventions into the wider ICU population is clearly essential. The panel believes that optimal use of existing therapies and the integration of proven new therapies will reduce mortality rates. Further positive results from new trials with improved trial designs should encourage intensivists to incorporate new interventions into their practice. Protocols are essential to ensure efficient integration of new therapies and to improve outcomes on the wards. Morris predicted in a recent paper that an increase in compliance with evidence-based recommendations through the use of protocols would decrease error and would enhance patient safety [49] . However, a complete treatment protocol is only effective when each ward (inside and outside of the ICU) has the trained staff to implement it, and when a skilled intensive care physician is available to lead the team. Training and education of staff is essential. All five of the interventions discussed in this article have generated convincing evidence for their use, and they hold out hope for reducing mortality in patients with sepsis, severe sepsis and septic shock. Yet, despite compelling data, the application of these interventions has yet to become routine practice in most ICUs. It is our hope that this article will enable physicians to understand how best to apply these therapies in clinical practice; from appropriate patient selection and timing of therapy, to combining different approaches for optimal patient management. A willingness to embrace new interventions, coupled with the development and implementation of rigorous protocols to ensure appropriate use, will improve outcomes and lead to a substantial reduction in mortality in these patients. • A respiratory rate ≥ 20 breaths/min or a partial pressure of arterial carbon dioxide ≤ 32 mmHg, or the use of mechanical ventilation for an acute respiratory process. • A white cell count ≥ 12,000/mm 3 or ≤ 4000/mm 3 , or a differential count showing >10% immature neutrophils. Patients should meet at least one of the following five criteria: • Pregnancy or breastfeeding. • Aged younger than 18 years or weight >135 kg. • Platelet count < 30,000/mm 3 . • Conditions that increase the risk of bleeding: • surgery requiring general or spinal anesthesia within 12 hours before the infusion, the potential need for such surgery during the infusion, or evidence of active bleeding postoperatively; • a history of severe head trauma requiring hospitalization, intracranial surgery, or stroke within 3 months before the study, or any history of intracerebral arteriovenous malformation, cerebral aneurysm, or mass lesions of the central nervous system; • a history of congenital bleeding diatheses; gastrointestinal bleeding within 6 weeks before the study unless corrective surgery had been performed; or • trauma considered to increase the risk of bleeding. • A known hypercoagualable condition including: • resistance to activated protein C; • hereditary deficiency of protein C, protein S, or antithrombin III; • presence of anticardiolipin antibody, antiphospholipid antibody, lupus anticoagulant, or homocysteinemia; or • recently documented (within 3 months) or highly suspected deep-vein thrombosis or pulmonary embolism. • Patient's family or physician, or both, not in favor of aggressive treatment of the patient, or the presence of an advanced directive to withhold life-sustaining treatment. • Patient not expected to survive 28 days because of an uncorrectable medical condition, such as poorly controlled neoplasm or other end-stage disease. • Moribund state in which death is perceived to be imminent. • Human immunodeficiency virus infection in association with a last known CD4 cell count ≤ 50/mm 3 . • History of bone marrow, lung, liver, pancreas, or smallbowel transplantation. • Chronic renal failure requiring hemodialysis or peritoneal dialysis (acute renal failure was not an exclusion criterion). • Known or suspected portosystemic hypertension, chronic jaundice, cirrhosis, or chronic ascites. • Acute pancreatitis with no established source of infection. • Participation in an investigational study within 30 days before treatment. • Use of any of the following medications or treatment regimens: • unfractionated heparin to treat an active thrombotic event within 8 hours before the infusion (prophylactic treatment with a dose of unfractionated heparin of up to 15,000 U/day was permitted); • low molecular weight heparin at a higher dose than recommended for prophylactic use (as specified in the package insert) within 12 hours before the infusion; • warfarin (if used within 7 days before study entry and if the prothrombin time exceeded the upper limit of the normal range for the institution); • acetylsalicylic acid at a dose of more than 650 mg/day within 3 days before the study; • thrombolytic therapy within 3 days before the study (thrombolytic agents permitted for the treatment of thromboses within a catheter); • glycoprotein IIb/IIIa antagonists within 7 days before study entry; • antithrombin III at a dose of more than 10,000 U within 12 hours before the study; • protein C within 24 hours before the study. Drotrecogin alfa (activated) should be given at a dose of 24 µg/kg/hour for 96 hours. Infusion should be interrupted 1 hour prior to any percutaneous procedure or major surgery, and should be resumed 1 and 12 hours later, respectively, in the absence of bleeding complications. There was an 8-hour time window from shock onset to check for eligibility and to perform a short ACTH test (blood samples before and 30 and 60 min after a 250 µg intravenous bolus of tetracosactrin). Patients were then randomly assigned to receive 50 mg hydrocortisone as an intravenous bolus every 6 hours and one 50 µg tablet of fludrocortisone through a nasogastric tube once a day, or their respective placebos. Treatments were given for 7 days, and patients were followed up for 1 year. On admission, patients should receive continuous intravenous glucose (200-300 g over 24 hours). After 24 hours, total parenteral, combined parenteral and enteral, or total enteral feeding should be instituted: 20-30 nonprotein kcal/kg/day with a balanced composition (0.13-0.26 g nitrogen/kg/day and 20-40% nonprotein calories in the form of lipids). Total enteral feeding should be attempted as early as possible.

Synergistic Roles of Eukaryotic Translation Elongation Factors 1Bγ and 1A in Stimulation of Tombusvirus Minus-Strand Synthesis

Host factors are recruited into viral replicase complexes to aid replication of plus-strand RNA viruses. In this paper, we show that deletion of eukaryotic translation elongation factor 1Bgamma (eEF1Bγ) reduces Tomato bushy stunt virus (TBSV) replication in yeast host. Also, knock down of eEF1Bγ level in plant host decreases TBSV accumulation. eEF1Bγ binds to the viral RNA and is one of the resident host proteins in the tombusvirus replicase complex. Additional in vitro assays with whole cell extracts prepared from yeast strains lacking eEF1Bγ demonstrated its role in minus-strand synthesis by opening of the structured 3′ end of the viral RNA and reducing the possibility of re-utilization of (+)-strand templates for repeated (-)-strand synthesis within the replicase. We also show that eEF1Bγ plays a synergistic role with eukaryotic translation elongation factor 1A in tombusvirus replication, possibly via stimulation of the proper positioning of the viral RNA-dependent RNA polymerase over the promoter region in the viral RNA template.These roles for translation factors during TBSV replication are separate from their canonical roles in host and viral protein translation.

Plus-stranded (+)RNA viruses recruit numerous host proteins to facilitate their replication and spread [1, 2] . Among the identified host proteins are RNA-binding proteins (RBPs), such as ribosomal proteins, translation factors and RNA-modifying enzymes [1] [2] [3] [4] [5] . The subverted host proteins likely affect several steps in viral RNA replication, including the assembly of the replicase complex and initiation of RNA synthesis. However, the detailed functions of recruited host RBPs in (+)RNA virus replication are known only for a small number of host factors [2, [6] [7] [8] . Tomato bushy stunt virus (TBSV) is model plant RNA virus coding for two replication proteins, p33 and p92 pol , which are sufficient to support TBSV replicon (rep)RNA replication in a yeast (Saccharomyces cerevisiae) model host [9, 10] . p33 and p92 pol are components of the membrane-bound viral replicase complex, which also contains the tombusviral repRNA serving not only as a template for replication, but also as a platform for the assembly of the viral replicase complex [11] [12] [13] . Recent genome-wide screens and global proteomics approaches with TBSV and a yeast host revealed a large number of host factors interacting with viral components or affecting TBSV replication. The identified host proteins are involved in various cellular processes, such as translation, RNA metabolism, protein modifications and intracellular transport or membrane modifications [14] [15] [16] [17] . Various proteomics analyses of the highly purified tombusvirus replicase has revealed at least five permanent resident host proteins in the complex, including the heat shock protein 70 chaperones (Hsp70) [18] [19] [20] [21] , glyceraldehyde-3-phosphate dehydrogenase [4] , pyruvate decarboxylase [21] , Cdc34p E2 ubiquitin conjugating enzyme [4, 21, 22] , eukaryotic translation elongation factor 1A (eEF1A) [23, 24] and two temporary resident proteins, Pex19p shuttle protein [25] and the Vps23p adaptor ESCRT protein [24, 26, 27] . The functions of several of these proteins have been studied in some detail [4, 17, 18, 19, 20] . The emerging picture from systems biology approaches is that eukaroyotic translation elongation factors (eEFs), such as eEF1A, play several roles during TBSV replication. Accordingly, eEF1A has been shown to facilitate the assembly of the viral replicase complex and stimulate the initiation of minus-strand synthesis by the viral RNA-dependent RNA polymerase (RdRp) [23, 24] . Another translation elongation factor identified in our genomewide screens with TBSV is eukaryotic elongation factor 1Bgamma (eEF1Bc) [15] . eEF1Bc is an abundant, but not essential cellular protein, which is part of the eukaryotic translation elongation factor 1B complex also containing the eEF1Ba subunit in yeast and the eEF1Ba and eEF1Bd subunits in metazoans [28] .The eEF1B complex is the guanine nucleotide exchange factor for eEF1A, which binds and delivers aminoacyl-tRNA in the GTPbound form to the elongating ribosome. Additional roles have been ascribed to eEF1Bc in vesicle-mediated intracellular protein transport, RNA-binding, vacuolar protein degradation, oxidative stress, intermediate filament interactions and calcium-dependent membrane-binding [29, 30, 31] . In this paper, we characterize the function of eEF1Bc in TBSV replication. Our approaches based on yeast and in vitro replication assays reveal that eEF1Bc is a component of the tombusvirus replicase and binds to the 39-end of the viral RNA. Using a cellfree replication assay, we define that eEF1Bc plays a role by enhancing minus-strand synthesis by the viral replicase. The obtained data support the model that eEF1Bc opens up a 'closed' structure at the 39-end of the TBSV (+)RNA, rendering the RNA compatible for initiation of (-)-strand synthesis. Moreover, we find that eEF1Bc and eEF1A play nonoverlapping functions to enhance (-)-strand synthesis. Altogether, the two translation factors regulate TBSV replication synergistically by interacting with different portions of the viral (+)RNA and the replication proteins. Deletion of eEF1Bc inhibits TBSV RNA accumulation in yeast model host eEF1Bc is coded by TEF3 and TEF4 nonessential genes in yeast [32, 33] . Single deletion of TEF3(CAM1) or TEF4 reduced TBSV repRNA accumulation to ,25% ( Figure 1A , lanes 3-8), while deletion of both genes resulted in even more inhibition, supporting TBSV repRNA accumulation only at 15% level (lanes 9-11). Expression of eEF1Bc (Tef4p) in tef4D yeast increased TBSV replication to ,80%, demonstrating that the defect in TBSV repRNA replication in tef4D yeast can be complemented.Altogether, these data established that eEF1Bc plays an important stimulatory role in TBSV replication. To obtain direct evidence on the involvement of eEF1Bc in TBSV replication, we prepared cell-free extracts (CFE) from a yeast strain lacking the TEF4 gene or from wt yeast. These yeast extracts contained comparable amount of total proteins ( Figure 1C , right panel). The CFE extracts were programmed with the TBSV (+)repRNA and purified recombinant p33 and p92 pol obtained from E. coli. Under these conditions, the CFE supports the in vitro assembly of the viral replicase, followed by a single cycle of complete TBSV replication, resulting in both (-)-stranded repRNA and excess amount of (+)-stranded progeny [20, 34] . Importantly in the case of a translation factor, this assay uncouples the translation of the viral proteins from viral replication, which are interdependent during (+)RNA virus infections. CFE obtained from tef4D yeast supported only 29% of TBSV repRNA replication when compared with the extract obtained from wt yeast ( Figure 1C , lane 2 versus 4). These data demonstrate that Tef4p plays an important role in the activity of the viral replicase complex. To test if the decrease in TBSV repRNA replication in vitro was due to reduced (+) or (-)-strand synthesis, we measured the replication products under non-denaturing versus denaturing conditions ( Figure 1C ). We found that the amount of dsRNA [representing the newly-synthesized 32 P-labeled (-)RNA product hybridized with the input (+)RNA; lane 1, Figure 1C , see also ref. [23] ] and the newly-synthesized (+)RNA both decreased by ,3fold in CFE obtained from tef4D yeast in comparison with those products in the wt CFE (lane 3). Since the ratio of dsRNA and ssRNA did not change much in the CFEs ( Figure 1C ), the obtained data are consistent with the model that Tef4p (eEF1Bc) affects the level of (-)RNA production, which then leads to proportionately lower level of (+)RNA progeny. Adding purified recombinant eEF1Bc to CFE from tef4D yeast supported TBSV repRNA replication to similar extent as the CFE from wt yeast (i.e., containing wt eEF1Bc, Figure 1D , lanes 3-6 versus 1-2), indicating that the recombinant eEF1Bc can complement the missing Tef4p in vitro, when the same amount of p33 and p92 pol was provided. Using large amount of eEF1Bc in the CFE-based assay did not further increase TBSV repRNA replication ( Figure 1D , lanes 3-4), suggesting that eEF1Bc should be present in optimal amount during TBSV replication. To obtain additional evidence if eEF1Bc could stimulate RNA synthesis by the viral RdRp, we used the E. coli-expressed recombinant p88C pol RdRp protein of Turnip crinkle virus (TCV). The TCV RdRp, unlike the E. coli-expressed TBSV p92 pol or the closely-related Cucumber necrosis virus (CNV) p92 pol RdRps, does not need the yeast CFE to be functional in vitro [35, 36] . Importantly, the template specificity of the recombinant TCV RdRp with TBSV RNAs is similar to the closely-related tombusvirus replicase purified from yeast or infected plants [10, 36, 37, 38] . The recombinant TCV RdRp preparation lacks co-purified eEF1Bc (E. coli does not have a homolog), unlike the yeast or plant-derived tombusvirus replicase preparations, facilitating studies on the role of eEF1Bc on the template activity of a viral RdRp. When we added various amounts of the highly purified recombinant eEF1Bc to the TCV RdRp assay programmed with TBSVderived SL3-2-1(+) RNA template, which is used by the TCV RdRp in vitro to produce the complementary (-)RNA product [37] , we observed a ,2-to-4-fold increase in (-)RNA synthesis by the TCV RdRp (Figure 2A , lanes [3] [4] [5] . eEF1Bc in the absence of the TCV RdRp did not give a 32 P-labeled RNA product, excluding that our eEF1Bc preparation contained RdRp activity (not shown). Altogether, our data suggest that eEF1Bc can stimulate in vitro activity of TCV RdRp on a TBSV (+)RNA template, confirming a direct role for eEF1Bc in viral (-)RNA synthesis by a viral RdRp. To test if the stimulating activity of eEF1Bc on the in vitro RdRp activity was due to binding of eEF1Bc to the (+)RNA template and/or to the TCV RdRp protein, we performed assays, in which the recombinant eEF1Bc was pre-incubated with the TCV RdRp or the (+)RNA template prior to the RdRp assay. These experiments revealed that pre-incubation of the purified eEF1Bc with the TBSV-derived SL3-2-1(+) RNA template prior to the RdRp assay led to a ,4.5-fold increase in (-)RNA products ( Figure 2B, lanes 1-2) . In contrast, pre-incubation of the TCV Author Summary RNA viruses recruit numerous host proteins to facilitate their replication and spread. Among the identified host proteins are RNA-binding proteins (RBPs), such as ribosomal proteins, translation factors and RNA-modifying enzymes. In this paper, the authors show that deletion of eukaryotic translation elongation factor 1Bgamma (eEF1Bc) reduces Tomato bushy stunt virus (TBSV) replication in a yeast model host. Knock down of eEF1Bc level in plant host also decreases TBSV accumulation. Moreover, the authors demonstrate that eEF1Bc binds to the viral RNA and is present in the tombusvirus replicase complex. Functional studies revealed that eEF1Bc promotes minusstrand synthesis by serving as an RNA chaperone. The authors also show that eEF1Bc and eukaryotic translation elongation factor 1A, another host factor, function together to promote tombusvirus replication. Cell-free TBSV replicase assay supports a role for eEF1Bc in minus-strand synthesis. Purified recombinant TBSV p33 (12 pmol) and p92 pol (1 pmol) replication proteins in combination with DI-72 (+)repRNA (4 pmol)were added to the whole cell extract prepared from tef4D (lanes 1-2) or WT yeast strains. Left panel: The nondenaturing PAGE analysis of the 32 P-labeled repRNA products obtained is shown. The full-length single-stranded repRNA is pointed at by an arrow. Odd numbered lanes represent replicase products, which were not heat treated (thus both ssRNA and dsRNA products are present), while the even numbered lanes show the heattreated replicase products (ssRNA is present). The amount of ssRNA and the ratio of ssRNA/dsRNA in the samples are shown. Note that, in the nondenatured samples, the dsRNA product represents the annealed (-)RNA and the input (+)RNA, while the ssRNA products represents the newly made (+)RNA products. Right panel shows the coomassie-blue stained SDS-PAGE gel to visualize total protein levels in the whole cell extracts. (D) eEF1Bc stimulates TBSV repRNA synthesis in whole cell extract prepared from tef4D. Increasing amounts of purified recombinant eEF1Bc (lanes 3-4, 26 pmol; lanes 5-6, 13 pmol) were added to tef4D CFE and the in vitro synthesized 32 P-labeled TBSV repRNA was measured on denaturing PAGE. See further details in panel C. Note that the recombinant eEF1Bc added to the tef4D CFE is about 10-fold less than the total eEF1Bc present in the WT CFE. doi:10.1371/journal.ppat.1002438.g001 RdRp with the (+)RNA template ( Figure 2B , lanes 3-4) or eEF1Bc with the TCV RdRp ( Figure 2B , lanes 7-8) prior to the RdRp assay did not result in increase in (-)RNA synthesis. Overall, data shown in Figure 2B imply that eEF1Bc can stimulate (-)RNA synthesis only when eEF1Bc binds to the (+)RNA template before the RdRp binding to the template. To further test the stimulatory effect of eEF1Bc, we also tested the RdRp activity in the presence of eEF1Bc using a mutated (+)RNA template. The mutation [SL3-2-1m(+)] opens up the closed structure in the promoter region that leads to increased template activity [39] . The mutated template showed only ,2-fold increased RNA products in the RdRp assay with eEF1Bc ( Figure 2C , lanes 3-4 versus 1-2). In contrast, eEF1Bc did not stimulate RNA products when the negative-stranded RI-III(-) RNA was used as a template in the TCV RdRp assay ( Figure 2C , lanes 9-10 versus 7-8). Thus, these data support the model that eEF1Bc can mainly stimulate (-)-strand synthesis by the RdRp on the wt 39 TBSV sequence, while it is not effective on the (-)RNA template. To test if eEF1Bc directly binds to a particular region within the TBSV repRNA, we performed electrophoretic mobility shift (EMSA) experiments with purified eEF1Bc and 32 P-labeled regions of (+)repRNA that included known cis-acting elements involved in (-)RNA synthesis [39, 40, 41] . These experiments revealed that eEF1Bc bound efficiently to the 39-end of the TBSV (+)repRNA (construct SL3-2-1, carrying the terminal 3 stem-loop structures, Figure S1 ). Template competition experiments confirmed that SL3-2-1 RNA bound competitively to eEF1Bc in vitro( Figure S1B ). To further define what sequence within SL3-2-1 is bound by eEF1Bc, we used complementary DNA oligos to partially convert portions of SL3-2-1 into duplexes (RNA/DNA hybrids) as shown The TCV RdRp assay had two steps: first, the shown components were incubated at room temperature to facilitate their interaction, followed 5 min later the addition of the shown component and the ribonucleotides to start RNA synthesis. The RdRp activity in samples containing the template RNA and the RdRp were chosen as 100% (lanes 5-6 and 9-10). The RNA transcript (20 pmol), eEF1Bc (20 pmol) and purified TCV RdRp (2 pmol) were used in these assays. (C) The effect of eEF1Bc on the TCV RdRp activity with additional templates. One of the templates was SL3-2-1 m(+) with a point mutation within the promoter sequence (carrying SL1m mutation), which is being used more efficiently than the wt SL3-2-1(+) by the TCV RdRp in vitro. The second template was RI-III ( in Figure 3A . EMSA assay with purified recombinant eEF1Bc revealed that the very 39-terminal SL1 region had to be ''free'' (not part of the duplex) for eEF1Bc to bind efficiently to the SL3-2-1 RNA (compare lane 1 with lane 5 in Figure 3A ). Since eEF1Bc is known to bind to A-rich single-stranded sequences [32] , we mutagenized the tetraloop (GAAA) sequence to either CUUG or GUUU tetraloop sequences ( Figure 3B ) that are expected to maintain the stability of the double-stranded stem. EMSA analysis showed that neither RNAs with the new tetraloop sequences bound efficiently to eEF1Bc ( Figure 3B , lanes 5-7 and 11-13). Based on the EMSA data, we conclude that the GAAA tetraloop region of SL1 is an efficient binding site for eEF1Bc in vitro. However, we cannot exclude that eEF1Bc binding may be dependent on stabilizing effects of the GNRA tetraloop on the stem structure. The loop nucleotides may or may not be involved in protein-RNA contacts. Binding of eEF1Bc to the 39 end of the TBSV RNA is required for stimulation of (-)-strand RNA synthesis in vitro To examine if binding of eEF1Bc to SL1 is important for stimulation of (-)-strand RNA synthesis by the viral RdRp, we performed an in vitro RNA synthesis assay using a mutated SL3-2-1 carrying the 'CUUG' tetraloop instead of the wt 'GAAA' tetraloop sequence ( Figure 4A ). Unlike for the wt SL3-2-1 RNA, eEF1Bc could not stimulate complementary RNA synthesis by the viral RdRp on the SL3-2-1cuug(+) template ( Figure 4A , lanes 7-10 versus 1-4). These data suggest that binding of eEF1Bc to the 'GAAA' tetraloop sequence of SL1 is important to stimulate (-)strand synthesis by the viral RdRp in vitro. Since the TBSV (+)RNA, including the minimal SL3-2-1 sequence, forms a secondary structure where the replication silencer sequence (RSE) in SL3 base-pairs with the 39-terminal 5 nts within the genomic promoter (gPR) (both sequences are highlighted with gray boxes in Figure 4A ), it is possible that eEF1Bc helps (-)-strand synthesis by opening up the gPR. The single-stranded gPR sequence would be more accessible for (-)strand synthesis as shown based on RNA mutagenesis [39] . To test this model, we obtained a complementary RNA that formed a duplex with SL1 and neighboring sequences, but leaving SL1 including the 'GAAA' loop-sequence nonbase-paired to facilitate binding to eEF1Bc ( Figure 4B ). Interestingly, eEF1Bc was able to stimulate (-)-strand synthesis by 70%, suggesting that eEF1Bc might indeed facilitate opening up the 39-terminal structure when it is part of a duplex. eEF1Bc co-purifies with the viral replicase complex and it binds to TBSV repRNA in yeast To test if eEF1Bc is a component of the tombusvirus replicase, we purified the His 6 -Flag-tagged p33 (HF-p33) replication protein via Flag-affinity purification from the detergent-solubilized membrane fraction of yeast [10] . We detected both p33 and eEF1Bc in the purified preparation ( Figure 5A , lane 1), suggesting that eEF1Bc is likely part of the replicase complex [21] . Importantly, eEF1Bc was not found in the control samples containing the His 6 -tagged p33 (H-p33) that were also purified via the Flag-affinity procedure ( Figure 5A , lane 2). Since eEF1Bc does not seem to bind to p33 or p92 replication proteins (data not shown), it is likely that eEF1Bc was co-purified with p33 via the viral RNA template in the viral replicase complex. To demonstrate that eEF1Bc can indeed bind to the TBSV (+)repRNA in cells, we Flag-affinity-purified His 6 -Flag-tagged eEF1Bc from the detergent-solubilized membrane fraction and also from the soluble (cytosolic) fraction of yeast. Interestingly, the viral RNA was co-purified with eEF1Bc from both fractions ( Figure 5B, lanes 3 and 7) . These data confirmed that eEF1Bc binds to the viral RNA in yeast. Since eEF1Bc was found in association with the TBSV repRNA in the cytosolic fraction of yeast, it is possible that eEF1Bc might affect the viral RNA recruitment from the cytosol into replication that takes place on the peroxisomal or ER membrane surfaces [42, 43] . Therefore, we tested the recruitment of the TBSV (+)repRNA to the membrane fraction in our CFE assay [23] . We found that eEF1Bc did not facilitate the association of the TBSV (+)repRNA with the membrane when applied in the absence of p33/p92 replication proteins ( Figure S2 ). Moreover, eEF1Bc did not further increase the amount of TBSV (+)repRNA bound to the membrane in the presence of p33/p92 replication proteins, which are needed for RNA recruitment ( Figure S2 , lanes 3-4 and 8-10) [24] . Therefore, we conclude that eEF1Bc is unlikely to promote the recruitment of the TBSV (+)repRNA to the membrane. Since both eEF1Bc and eEF1A bind to the 39-terminal region of the TBSV (+)RNA ( Figure 3 ) and ref: [23, 24] , it is possible that they could affect each other's functions during replication. To test the mutual effect of eEF1Bc and eEF1A on the (-)-strand RNA production of the viral RdRp, we performed in vitro RdRp assays with purified eEF1A and recombinant eEF1Bc as shown in Figure 6 . Based on previous experiments, eEF1Bc was known to stimulate (-)-strand synthesis the most when pre-incubated with the template (+)RNA ( Figure 2B ). In contrast, pre-incubation of eEF1A with the viral RdRp was more effective than preincubation of eEF1A with the template RNA [23] . Therefore, we performed the pre-incubation experiments prior to the RdRp assay as shown in Figure 6 . We found the largest stimulation of (-)-strand synthesis by the viral RdRp in a dual pre-incubation assay, when eEF1Bc was pre-incubated with the viral RNA template, while eEF1A was separately pre-incubated with the viral RdRp ( Figure 6, lanes 3-4) . Pre-incubation of eEF1Bc with the viral RNA template (lanes 5-6) or pre-incubation of eEF1A with the viral RdRp (lanes 7-8) were about half as efficient in stimulation of (-)-strand synthesis than the dual pre-incubation assay (lanes [3] [4] . Therefore, these data support the model that eEF1Bc and eEF1A both promote (-)-strand synthesis and their effect is synergistic, likely involving separate mechanisms (see Discussion). in vitro binding assay with purified eEF1Bc using an ssDNA oligo/ssRNA template duplex. The annealed ssDNA (purple)/ssRNA (black) duplexes representing the 39 end of the TBSV RNA are shown schematically. The assay contained the annealed ssDNA/ssRNA plus 0.6 and 0.4 pmol purified recombinant eEF1Bc, respectively. The 32 P-labeled free ssDNA and ssDNA/ssRNA duplex were separated on nondenaturing 5% acrylamide gels. Quantification of the ssDNA/ssRNA duplex was done with ImageQuant. (B) RNA gel shift analysis shows the role of the SL1 tetraloop in binding to eEF1Bc. The RNA templates representing the 39 end of the TBSV RNA and the mutations (circled nucleotides) are shown schematically. The eEF1Bc -32 P-labeled ssRNA complex was visualized on nondenaturing 5% acrylamide gels. The RNA transcript (0.2 pmol), and eEF1Bc (0.4, 0.5 and 0.6 pmol) were used in these assays. doi:10.1371/journal.ppat.1002438.g003 To obtain evidence on the importance of eEF1Bc in TBSV replication in the natural plant hosts, we knocked down the expression of the eEF1Bc gene in Nicotiana bethamiana leaves via VIGS (virus-induced gene silencing). Efficient knocking down of eEF1Bc mRNA level in N. benthamiana ( Figure 7B ) only resulted in slightly reduced growth of the plants without other phenotypic effects ( Figure 7A ). The accumulation of TBSV genomic RNA, however, was dramatically reduced in both inoculated ( Figure 7B , lanes 1-5) and the systemically-infected young leaves ( Figure 7C , lanes 1-4) when compared with the control plants infected with the 'empty' Tobacco rattle virus (TRV) vector. The lethal necrotic symptoms caused by TBSV in N. benthamiana were also greatly attenuated in the eEF1Bc knock-down plants ( Figure 7A ). Therefore, we conclude that eEF1Bc is essential for TBSV genomic RNA accumulation in N. bethamiana. To test if eEF1Bc is also needed for the replication of other plant RNA viruses, we infected eEF1Bc-silenced N. benthamiana leaves with the unrelated Tobacco mosaic virus (TMV) RNA ( Figure 8A ). We found that the severe symptoms caused by TMV were greatly ameliorated in eEF1Bc knock-down plants ( Figure 8A ). Accumulation of TMV genomic RNA was also dramatically reduced in both inoculated ( Figure 8B ) and systemically-infected ( Figure 8C ) leaves of the eEF1Bc knock-down plants. Based on these data, eEF1Bc seems to be needed for TMV replication and/or spread in plants. Thus, our data have revealed new functions for eEF1Bc in plant RNA virus replication and spread. Tombusviruses, similar to other (+)RNA viruses, subvert a yet unknown number of host-coded proteins to facilitate robust virus replication in infected cells. The co-opted host proteins could be part of the viral replicase complexes and provide many yet undefined functions. Translation factors, such as eEF1Bc and eEF1A, are among the most common host factors recruited for (+)RNA virus replication [23, 24] . While eEF1A is an integral component of the tombusvirus replicase complex [23, 24] and several other viral replicases [44, 45, 46] , the function of eEF1Bc in tombusvirus replication is studied in this paper. Co-purification experiments with the p33 replication protein, which is the most abundant protein component in the tombusvirus replicase complex [21, 22] , revealed that eEF1Bc is a permanent member of the replicase ( Figure 5A ). eEF1Bc is likely recruited into the viral replicase via the viral (+)RNA, which is bound to eEF1Bc in both cytosolic and membranous fractions ( Figure 5B ). The possible role of host proteins or membrane lipids in assisting the recruitment of eEF1Bc for TBSV replication cannot be excluded. Accordingly, eEF1Bc has been shown to bind to a large number of host proteins (www.yeastgenome.org). For example, eEF1A, which is also a permanent member of the tombusvirus replicase, is known to interact with eEF1Bc [47, 48, 49] and eEF1A might facilitate the recruitment of eEF1Bc and possibly other translation factors. The binding of eEF1Bc to intracellular membranes has also been shown before [32] . Altogether, our model predicts that the viral (+)RNA could be involved in recruitment of eEF1Bc into viral replication ( Figure 5) . However, the opposite model that eEF1Bc facilitates the recruitment of the TBSV (+)RNA into replication is not supported by our in vitro data ( Figure S2) . Indeed, addition of eEF1Bc to the CFE assay did not increase the membrane-bound fraction of TBSV (+)repRNA in the absence or presence of the viral replication proteins ( Figure S2 ). eEF1Bc selectively enhances minus-strand synthesis by opening the closed 39-terminus during TBSV RNA replication We confirmed a direct role for eEF1Bc in RNA synthesis in vitro by using a cell-free extract prepared from tef4D yeast that supported (-)-strand RNA synthesis ,3-fold less efficiently than CFE from wt yeast (Figure 1 ). Moreover, in vitro assays with highly purified eEF1Bc and the recombinant TCV RdRp, which is closely homologous with the TBSV p92 pol , also revealed that eEF1Bc stimulates (-)-strand synthesis by binding to the viral (+)RNA template ( Figure 3) . Accordingly, pre-incubation of eEF1Bc and the TBSV-derived template RNA prior to the RdRp assay led to the highest level of stimulation of (-)RNA synthesis ( Figure 2 ). On the other hand, eEF1Bc does not stimulate the RdRp activity directly, since pre-incubation of eEF1Bc with the RdRp did not lead to more efficient (-)-strand RNA synthesis in vitro ( Figure 2 ). We propose that eEF1Bc modifies the structure of the (+)-strand template prior to initiation of (-)-strand synthesis that leads to more efficient RNA synthesis as described below. In vitro initiation of (-)-strand synthesis by the viral RdRp requires the gPR promoter consisting of a short 39-terminal singlestranded tail and a stem-loop (SL1) sequence [39, 50] . However, Figure 6 . Synergistic effect of eEF1Bc and eEF1A on stimulation of minus-strand synthesis by the closely-related TCV RdRp. Purified eEF1Bc (20 pmol) and eEF1A (20 pmol) were added to the TCV RdRp (2 pmol) assay as shown. The RdRp assay had two steps: first, the shown components on the top and bottom were incubated in separate tubes at room temperature to facilitate their interaction, followed 5 min later by mixing the components from the two tubes and addition of the ribonucleotides to start RNA synthesis. The RdRp activity in samples containing the template RNA and the RdRp were chosen as 100% (lanes 1-2 and 9-10). The gel image shows the results of RNA synthesis in the presence of equal amounts of purified eEF1Bc and eEF1A as shown in a TCV RdRp assay. doi:10.1371/journal.ppat.1002438.g006 The FLAG/His 6 -tagged HF-p33 was purified from yeast extracts using a FLAG-affinity column. The purified HF-p33 and the co-purified His 6 -tagged eEF1Bc were detected with anti-His antibody. Bottom panel: Western blot of HF-p33 and the His 6 -tagged eEF1Bc in the total yeast extract using anti-His antibody. (B) RT-PCR analysis to detect the co-purified TBSV (+)RNA in the affinity-purified His 6 -tagged eEF1Bc preparation from yeast replicating TBSV repRNA. Both the membrane and soluble yeast fractions were used for eEF1Bc purification and subsequent RT-PCR analysis to detect (+)repRNA. ''+'' and ''-'' mean that His 6 -tagged eEF1Bc was expressed from a plasmid or not in yeast. Samples were used for RT-PCR (lanes 3-4 and 7-8) or for PCR (without RT reaction, lanes 1-2 and 5-6). doi:10.1371/journal.ppat.1002438.g005 the gPR region is present in a 'closed' structure in the TBSV (+)RNA due to base-pairing of a portion of the gPR with the RSE present in SL3 as shown in Figure 9 . This interaction makes the TBSV (+)RNA poor template in the in vitro assay due to the difficulty for the viral RdRp to recognize and/or open the 'closed' structure [39] . Our current work with eEF1Bc, however, suggests that eEF1Bc can bind to the tetraloop region of SL1 (and to an Arich sequence in SL2) that leads to melting of the base-paired structure and opening the stem of SL1 and the RSE-gPR basepairing as shown schematically in Figure 9B . We propose that the open structure can be recognized efficiently by the viral replicase leading to efficient initiation of (-)-strand synthesis ( Figure 9B ). This model is supported by several pieces of evidence presented in this paper, including (i) stimulation of (-)-strand synthesis by eEF1Bc when the wt SL1 is present in the template; (ii) lack of stimulation of(-)-strand synthesis by eEF1Bc when a mutated SL1 (tetraloop mutant), which does not bind efficiently to eEF1Bc, was used as a template in the in vitro assay; (iii) stimulation of (-)-strand synthesis when eEF1Bc was pre-incubated with the (+)-strand template, but not when eEF1Bc was pre-incubated with the viral RdRp ( Figure 2) ; and (iv) the lack of stimulation of (+)-strand synthesis on a (-)-strand template by eEF1Bc (Figure 2 ). In addition, eEF1Bc stimulated (-)-strand synthesis by the viral RdRp when a partially complementary RNA oligo was hybridized with the SL1 region ( Figure 4B ). However, eEF1Bc could not efficiently bind to the 39-end of the TBSV RNA when it formed a hybrid (duplex) with a perfectly complementary DNA oligo ( Figure 3A) , suggesting that eEF1Bc can melt only the local secondary structure, but cannot unwind more extended duplex regions. An alternative possibility is that eEF1Bc protein stabilizes the unpaired structure (when the SL1 structure is kinetically pairing/unpairing), rather than implying that it actively "opens" the structure. An intriguing aspect of our model is the possible regulation of the ''open'' and ''closed'' structure of the 39 UTR by eEF1Bc. Displacement of eEF1Bc bound to the 39-end by the viral replicase during (-)-strand synthesis could make the 39-terminus of the (+)strand RNA fold back into a 'closed' structure. This could prevent efficient re-utilization of the original (+)-strand template during TBSV replication, and the switch to efficient (+)-strand synthesis on the (-)RNA intermediate ( Figure 9B ). This model can also explain why the newly made (+)-strand RNA progeny will not enter the replication cycle in the absence of bound eEF1Bc within the originally-formed replicase complexes as observed previously in the CFE assay [20] . We propose that the new (+)RNA progeny need to leave the replicase complex, then bind to eEF1Bc in the cytosol and assemble new replicase complexes, followed by a new round of viral RNA replication. Thus, this model suggests that eEF1Bc plays a key role in regulation of the use of (+)-strand RNAs in TBSV replication ( Figure 9B ). Our finding of TBSV RNA binding by eEF1Bc adds to the growing list of RNAs bound by eEF1Bc. For example, the 39 UTR of vimentin mRNA is bound by eEF1Bc [51] , which led the authors to suggest that eEF1Bc plays a role in vimentin mRNA subcellular localization by also binding to cytoskeleton or membranes. eEF1Bc also binds to the tRNA-like structure at the 39 UTR of BMV, albeit the relevance of this binding is currently unclear [51] . Also, the actual role of eEF1Bc in the VSV replicase is currently not defined [31] . Translation elongation factors seem to be important for replication of many RNA viruses. For example, EF-Tu and EF-Ts play a role in replication of bacteriophage Qbeta [52, 53] . The eukaryotic homolog of EF-Tu, eEF1A was found to bind to viral RNAs, such as TBSV, Turnip yellow mosaic virus (TYMV) [54] , West Nile virus (WNV), Dengue virus, hepatitis delta virus, TMV, Brome mosaic virus, and Turnip mosaic virus [55, 56, 57, 58, 59, 60] and to viroid RNAs [61] . Therefore, it is highly probable that many (+)strand RNA viruses recruit translation elongation factors to facilitate and regulate their replication in infected cells. Nonoverlapping roles of eEF1Bc and eEF1A in stimulation of (-)-strand synthesis The emerging picture on the functions of eEF1Bc and eEF1A is that these translation elongation factors play different, yet complementary roles in TBSV replication as suggested in Figure 9B . While eEF1Bc binds to SL1, eEF1A has been shown to bind to both p92 pol RdRp and the SL3 region of TBSV (+)repRNA [23, 24] . The binding of the RNA by eEF1Bc promotes the opening of the closed 39-terminal structure, whereas eEF1A facilitates the proper and efficient binding of the RdRp to the 39 terminal RSE sequence of the viral RNA, which is required for the assembly of the viral replicase complex [11, 39] , prior to initiation of (-)-strand synthesis (Figure 9 ) [23, 24] . The binding of eEF1A-RdRp complex to the RSE might lead to proper positioning of the RdRp over the 39-terminal gPR promoter sequence opened up by eEF1Bc, thus facilitating the initiation of (-)RNA synthesis starting from the 39-terminal cytosine ( Figure 9B) . Altogether, the two translation factors facilitate the efficient initiation of (-)-strand synthesis in addition to reducing the possibility of re-utilization of the (+)-strand template for additional rounds of (-)-strand synthesis. This regulation of RNA synthesis by the co-opted host factors shows the specialized use of host components to serve the need of viral replication. The current work also provides evidence that eEF1Bc is a key factor in TBSV replication in yeast ( Figure 1 ) and in N. benthamiana (Figure 7) . Since eEF1Bc is a highly conserved protein in all eukaryotes [32] , it is not surprising that yeast eEF1Bc, similar to the plant eEF1Bc, can be co-opted for TBSV replication. Interestingly, deletion of either TEF3 or TEF4 genes reduced TBSV repRNA accumulation in yeast, suggesting that eEF1Bc is present in limiting amount or eEF1Bc is present in not easily accessible forms (in protein complexes) and/or locations in yeast cells. Silencing of eEF1Bc in N. bethamiana showed even more inhibition of TBSV RNA accumulation than deletion of eEF1Bc genes in yeast. This is likely due to the robust antiviral response (i.e., induced gene silencing) of the plant host, which could result in degradation of the small amount of viral RNA produced by the less efficient viral RNA replication in the presence of limited eEF1Bc in the knock-down plants. Silencing of eEF1Bc expression in N. benthamiana also reduced the accumulation of the unrelated TMV (Figure 8 ), which belongs to the alphavirus-like supergroup. These data suggest that eEF1Bc is likely involved in TMV replication, which also contains a highly structured 39-end [54] . Therefore, it is possible that eEF1Bc is co-opted by different plant RNA viruses, and possibly other RNA viruses as well. Overall, the current work suggests three major functions for eEF1Bc in TBSV replication ( Figure 9 ): (i) enhancement of the minus-strand synthesis by opening the 'closed' 39-end of the template RNA; (ii) reducing the possibility of re-utilization of (+)strand templates for repeated (-)-strand synthesis; and (iii) in coordination with eEF1A, stimulation of the proper positioning of the viral RdRp over the promoter region in the viral RNA template. These roles for eEF1Bc and eEF1A are separate from their canonical roles in host and viral protein translation. Saccharomyces cerevisiae strain BY4741 (MATa his3D1 leu2D0 met15D0 ura3D0) and the single-gene deletion strain of the TEF4-encoded form of eEF1Bc (tef4D) were obtained from Open Biosystems (Huntville, AL). TKY680 strain in which both yeast encoded eEF1Bc, TEF4 and TEF3 were deleted (MATa ura3-52 leu2D1 his3D200 trp1D101 lys2-801 tef3::LEU2 tef4::TRP1) and its isogenic wild type TKY677 (MATa ura3-52 leu2D1 his3D200 trp1D101 lys2-801) as well as the isogenic single deletion mutant strains, TKY678 (MATa ura3-52 leu2D1 his3D200 trp1D101 lys2-801 tef3::LEU2) and TKY 679 (MATa ura3-52 leu2D1 his3D200 trp1D101 lys2-801 tef4::TRP1) were published previously [30] . The following plasmids pESC-GAL1-Hisp33/GAL10-DI-72, pGAD-CUP1-p92 pYES-GAL1-p92, pCM189-TET-His92 were described earlier [21, 22] . URA3 based pGBK-ADH-Hisp33/ GAL1-DI72, pGBK-CUP1-HisFLAGp33/GAL1-DI-72, and pGBK-CUP1-Hisp33/GAL1-DI-72 plasmids were constructed by Daniel Barajas (unpublished result). The URA3 based, low copy-number plasmid, pYC-GAL1-Tef4 expressing non-tagged full-length Tef4 protein was constructed as follows: pYC/NT-C plasmid was digested with BamHI and XhoI restriction enzymes and then PCR product of the TEF4 gene was generated with primers #2089 (ccgcGGATCCATGTCCCAAGGTACTTTA-TAC) and #2320 (CGCCTCGAGTTATTTCAAAACCT-TACCGTCAACAATTTCC) and digested with the same restriction enzymes, followed by ligation. The plasmid pYES-NTC2-GAL1-HisTef4 expressing His 6 -tagged Tef4p protein was created with the same restriction enzymes using pYES-NT-C2. HIS3-based pEsc-His/Cup-FLAG plasmid [20] was digested with BamHI and XhoI restriction enzymes and then PCR product of the TEF4 gene was generated with primers #2089 and #2320 and digested with the same restriction enzymes, followed by ligationto obtain pEsc-His/Cup-FLAG-TEF4. HIS3 based pESC-GAL1-His33/GAL10-DI-72 and LEU2 based pGAD-CUP1-Hisp92 plasmids were transformed into tef4D strain. In the in vivo complementation assay, non-tagged Tef4p protein was expressed from URA3 plasmid pYC-GAL1-Tef4 and TEF4 mRNA was detected with a specific probe generated by the T7 transcription of the PCR product obtained with primers #2089 and #3788 (TAATACGACTCACTATAGGATTATT-TCAAAACCTTACCGTCAACAATTTCC). TKY680 (tef3D/tef4D), the isogenic TKY679 (tef4D), TKY678 (tef3D) and wild type TKY677 yeast were transformed with plasmids pESC-GAL1-His33/GAL10-DI-72 and pCM189-TET-His92. Yeast was pre-grown at 23uC overnight in 3 ml synthetic complete dropout medium lacking the relevant amino acids containing 2% glucose and 1 mg/ml doxycyclin to suppress p92 expression by the inhibition of TET promoter and then TBSV replication was launched by replacing the media with 2% galactose without doxycycline. Cells were harvested at 48 h time point. Total RNA extraction from yeast cells and Northern blotting and Western blotting were done as previously described [15, 24] . Expression and purification of recombinant eEF1Bc protein pEsc-His/Cup-FLAG-TEF4 plasmid was transformed into tef4D strain. Yeast was pre-grown overnight at 29uC in 2 ml synthetic complete dropout medium lacking histidine (SC-Hmedium) containing 2% glucose. The volume of the media was increased up to 100 ml 16 h later and copper sulfate was added to a final concentration of 50 mM for induction of protein expression. Yeast was grown to 0.8 OD 600 (,4-6 h). Then, yeast cells were harvested and broken by glass beads in a FastPrep cell disruptor followed by Flag-affinity purification of FLAG-Tef4p protein [34] . The bacterial heterologous expression and purification of His 6tagged Tef3 protein from plasmid pTKB523 was performed as described in ref: [62] using only the Ni affinity column step. Yeast extract capable of supporting TBSV replication in vitro was prepared as described [20] . The newly synthesized 32 P-labeled RNA products were separated by electrophoresis in a 5% polyacrylamide gel (PAGE) containing 0.5x Tris-borate-EDTA (TBE) buffer with 8 M urea. To detect the double-stranded RNA (dsRNA) in the cell-free replication assay, the 32 P-labeled RNA samples were divided into two aliquotes: one half was loaded onto the gel without heat treatment in the presence of 25% formamide, while the other half was heat denatured at 85uC for 5 min in the presence of 50% formamide [20] . To test the in vitro activity of Tef4p, different concentrations (26 and 13 pmol) of purified FLAG/His 6 -Tef4p was added to 0.25 mg (4 pmol) DI-72 (+)repRNA transcript and incubated in the presence of yeast cell-free extract and reaction buffer for 10 minutes at RT followed by the addition of MBP-p33 and MBP-p92 along with the rest of the reaction components. The reaction was performed at 25uC for 3 h and analyzed as above. The TCV RdRp reactions were carried out as previously described for 2 h at 25uC [36] , except using 7 pmol template RNA and 2 pmol affinity-purified MBP-p88C. Different concentrations of eEF1Bc (6xHis-affinity purified recombinant Tef3p obtained from E. coli or Flag-affinity purified HF-Tef4p obtained from yeast) were added to the reaction at the beginning or as indicated in the text and Figure 2 . legend. The 32 P-labeled RNA products were analyzed by electrophoresis in a 5% PAGE/8 M urea gel [63] . The 86-nt 39 noncoding region of TBSV genomic RNA and its mutants were used as the template in the RdRp assay [24, 36] . RNA templates were generated with T7 transcription using PCR products obtained with the following primers: #1662 (TAATACGACTCACTATAG GACACG GTTG ATC TC ACC-CTTC) and #1190 (GGGCTGCATTTCTGCAATG) for SL3-2-1(+), #1662 and #4390 (GGGCTGCACAAGTGCAAT-GTTCCGGTTGTCCGGT) for SL3-2-1cuug(+). SL3-2-1m(+) RNA was generated with T7 transcription on PCR products amplified with primers #1662 and #1190, on a plasmid template harboring GGGCU nucleotide-deletion in SL3 region as described [39] . A duplex RNA was generated by hybridizing SL3-2-1(+) and SL3-2-ds1(-) made by T7 transcription of the PCR product using primers #4361 (GTAATACGACTCACTA-TAGGGCTACTTCCGGTTGTCCGGTAGTGCTTCC) and For EMSA, 6xHis-Flag tagged Tef4p was purified from a yeast tef4D strain with anti-FLAG M2-agarose affinity resin. Different concentrations (0.6, 0.5 and 0.4 pmol) of HF-Tef4p protein was used for incubation with 0.2 pmol of 32 P-labeled SL3/2/1(+) RNA or mutated RNAs at 25uC in a binding buffer [50 mM Tris-HCl (pH 8.2), 10 mM MgCl 2 , 10 mM DTT, 10% glycerol, 2 U of RNase inhibitor (Ambion)]. Samples were incubated at 25uC for 15 min, then resolved in 4% nondenaturing polyacrylamide gel [23] . Similar experiments were also performed with 6xHis-affinity purified recombinant Tef3p obtained from E. coli (not shown). For the co-purification of TBSV DI-72 repRNA and eEF1Bc protein, the yeast tef4D strain was co-transformed with pGBK-ADH-Hisp33/GAL1-DI72, pGAD-CUP1-Hisp92 and pESC-CUP1-HisFLAG-Tef4. The pESC-CUP1-FLAGHis-Tef4 plasmid was replaced with the pESC plasmid in the control experiment. Yeast was pre-grown overnight at 29uC in 2 ml SC ULHmedium containing 2% glucose and 5 mM copper sulfate. The volume of the media was increased to 20 ml after 16 h for an additional 10 h (OD 600 of ,0.8), then the cultures were transferred to 20 ml SC ULHmedium containing 2% galactose to induce TBSV DI-72 RNA transcription at 23uC. The transcription of DI-72 RNA was stopped by changing to the media containing 2% glucose after 8 h. The cultures were diluted to 200 ml and copper sulfate was added to a final concentration of 50 mM to induce the expression of Flagtagged Tef4 protein. After incubation at 23uC for 24 h, the samples were centrifuged at 3000 rpm for 4 min. Cells (,1 g) were re-suspended in 2 ml TG Buffer (50 mM Tris-HCl [pH 7.5], 10% glycerol, 15 mM MgCl 2 , and 10 mM KCl) supplemented with 0.5 M NaCl and 1% [V/V] YPIC yeast protease inhibitor cocktail (Sigma) and RNase inhibitor (Ambion). Yeast cells were broken by glass beads in a FastPrep cell disruptor (MP Biomedicals) 4 times for 20 sec each at speed 5.5. Samples were removed and incubated 1 min in an ice-water bath after each treatment. The samples were centrifuged at 500 6g for 5 min at 4uC to remove glass beads, unbroken cells and debris then supernatant was moved into fresh pre-chilled tubes. After being centrifuged again at 500 6g for 5 min at 4uC supernatant transferred into fresh pre-chilled tubes and soluble (SU) and membrane (ME) fractions containing the viral replicase complex were separated with centrifugation at 35,000 6g for 15 min at 4uC. The SU fraction was applied on 0.1 ml anti-FLAG M2agarose affinity resin (Sigma) and Tef4 protein tagged with 6xHisand FLAG affinity tags was purified. Before applying ME fraction on the anti-FLAG M2 resin, solubilization of the membranebound replicase was performed in 1 ml TG buffer with 0.5 M NaCl, 1% [V/V] YPIC yeast protease inhibitor cocktail (Sigma), and 2% Triton X-100 via rotation for 2 hours at 4 uC. The solubilized membrane fraction was centrifuged at 35,000 6g at 4uC for 15 min and the supernatant was added to the resin preequilibrated with TG buffer supplemented with 0.5 M NaCl and 0.5% Triton X-100, followed by gentle rotation for 2 h at 4uC. The unbound proteins were removed by gravity flow, and the resin was washed two times with 1 ml TG buffer supplemented with 0.5 M NaCl, 0.5% Triton X-100 and once with 1 ml TG buffer, 0.5% Triton without NaCl. The bound proteins were eluted with 150 ml TG buffer without NaCl, 0.5% Triton X-100, supplemented with 150 mg/ml flag peptide and 1% yeast protease inhibitor cocktail via gentle tapping the column occasionally for 2 h at 4uC. After centrifugation at 600 6g 2 min at 4uC, semiquantitative RT-PCR was performed to detect TBSV repRNA copurified with eEF1Bc using primers, #359 (GTAATACGACT-CACTATAGGAAATTCTCCAGGATTTC) and #1190, amplifying full length (+)repRNA. To test if eEF1Bc is present in the viral replicase, yeast tef4D strain was transformed with pGBK-CUP1-HisFLAGp33/GAL1-DI-72, pGAD-CUP1-Hisp92 and pYES-GAL1-HisTef4. In the control experiment, 6xHisp33was expressed from pGBK-CUP1-Hisp33/GAL1-DI-72. Yeast cultures were grown in SC-ULHmedia containing 1% raffinose and 1% galactose with 5 mM copper-sulfate for 4 days with increasing the volume of the culture from 2 ml to 100 ml to a final OD 600 of, 1.0. After harvesting of cells, co-purification of 6xHis-tagged Tef4p with HF-p33 (part of the viral replicase) was conducted by using anti-FLAG M2-agarose affinity resin as described above (in the section: FLAG-affinity purification of eEF1Bc-TBSV repRNA complex), with the exception that only solubilized ME fraction was loaded on the column. Proteins bound to affinity resin were eluted by incubation with 150 ml buffer containing FLAG peptide and precipitated with Trichloroacetic acid (TCA) [64] . Samples were analyzed by SDS-PAGE and Western blotting. Virus-induced gene silencing (VIGS) in N. benthamiana was done as described [65, 66] . To generate the VIGS vector (pTRV2-eEF1BcNt), a 314-bp cDNA fragment of NteEF1Bc was RT-PCR amplified from a total RNA extract of N. benthamiana using the following pair of primers: #2993 (CGCGGATCCAAAG-GTTTCTGGGACATGTATGA) and #2994 (CGCCTCGA-GACACGCTCCTTCTGTGATTCATC) and inserted into the corresponding (BamHI/XhoI) restriction sites of pTRV2 plasmid. The sequence of the N. tabacum eEF1Bc gene (GenBank: ACB72462.1) was derived via a BLASTP search based on the Cterminal (translation elongation factor) domain (aa 252-412) of the Saccharomyces cerevisie Tef4 protein. The selected sequence (TC64920) from the Solanaceae Genomics Resource (www.tigr. org) gave 98% identity with N. tabacum EF1Bc -like gene (GB#: EU580435.1). To confirm the silencing of the EF1Bc gene in N. benthamiana, we performed RT-PCR amplification with primer pairs: #2952 (CGCGGATCCGGAAAGGTTCCTGTGCTTGA) and #2992 (C G C CT C G A G GTCCAGAAGTATCTCTCTACA TGTGG) on total RNA extract of pTRV2-EF1BcNt and pTRV2 empty agroinfiltrated N benthamiana plants. PCR conditions were as follows: 27 cycles of 94uC 20sec, 60uC 30sec, 68uC 30 sec with HiFi Taq polymerase. Tubulin mRNA control from the same total RNA samples was detected by RT-PCR using primers #2859 (TAATACGACTCACTATAGgaACCA AA TC AT T CATGTT-GCTCTC) and #2860 (TAGTGTATGTGATATCCCACCAA) [65] . The leaves of VIGS-treated plants were sap inoculated with TBSV, or TMV on the 9 th day after silencing [65] . Total RNA was extracted 3 or 5 days post inoculation [65] . For Northern blot analysis of the viral RNA level, we prepared 32 P-labeled complementary RNA probes specific for the 39-ends of the viral genomic RNAs based on T7 transcription. To obtain the PCR templates for the probes, we used the following primers for TBSV: #1165 (AGCGAGTAAGACAGACTCTTCA) and #22; for TMV: #2890 (TCTGGTTTGGTTTGGACCTC) and #2889 (GTAATACGACTCACTATAGGGATTCGAACCCC-TCGCTTTAT). The TBSV viral RNA is recruited to the membrane from the soluble fraction with the help of TBSV replication proteins and host factors present in the yeast CFE. The in vitro RNA recruitment reaction was performed according to [20, 23] , except that 32 Plabeled DI-72 (+)repRNA were used and rCTP, rUTP, 32 P-labeled UTP, and Actinomycin D were omitted from the assay. As a negative control, p33 and p92 were omitted from the reaction to detect DI-72 binding nonselectively to host proteins present in the membrane. In vitro binding assay with purified recombinant eEF1Bc (Tef3). The TBSV (+)RNA templates were the four noncontiguous segments of the TBSV (+)RNA that are present in defective interfering RNAs, including DI-72 repRNA used in this study. RI(+) represents the 59-UTR, RII(+)-SL is an internal highly conserved sequence that binds to p33 replication protein, RIII(+) is ashort conserved sequence closed to the 39 end, andSL3-2-1(+), which contains the promoter region (SL1) for initiation and the replication silencer element (within SL3) that down-regulates initiation. The assay contained 32 P-labeled free ssRNA (as shown), plus 0.6 pmol purified recombinant eEF1Bc, respectively. The bound RNA-protein complexes were separated on nondenaturing 5% acrylamide gels. Quantification of the free (unshifted) RNA was done with ImageQuant. (B) RNA gel shift analysis shows SL3-2-1(+) RNA binds competitively to eEF1Bc. The RNA templates representing the 39 end of the TBSV RNA and the deleted nucleotides are shown schematically. The cold competitor was SL3-2-1(+) RNA, which represents a large portion of the 39-UTR ( Figure 4A ). The eEF1Bc -32 P-labeled ssRNA complex was visualized on nondenaturing 5% acrylamide gels. (EPS) Figure S2 eEF1Bcdoes not affect the template recruitment step in vitro. Purified recombinant p33/p92 and 32 P-labeled DI-72 (+)repRNA and eEF1Bc (affinity purified recombinant Tef3) were added to a whole cell extract (CFE), followed by centrifugation/ washing to remove the 32 P-labeled repRNA that is not bound to the membrane. Then the membrane-bound RNA was analyzed in a denaturing PAGE gel. Note that the repRNA binds to the cellular membrane fraction nonspecifically (,20% level) in the absence of the viral replication proteins. (EPS)

Mannose-Binding Lectin Contributes to Deleterious Inflammatory Response in Pandemic H1N1 and Avian H9N2 Infection

Background. Mannose-binding lectin (MBL) is a pattern-recognition molecule, which functions as a first line of host defense. Pandemic H1N1 (pdmH1N1) influenza A virus caused massive infection in 2009 and currently circulates worldwide. Avian influenza A H9N2 (H9N2/G1) virus has infected humans and has the potential to be the next pandemic virus. Antiviral function and immunomodulatory role of MBL in pdmH1N1 and H9N2/G1 virus infection have not been investigated. Methods. In this study, MBL wild-type (WT) and MBL knockout (KO) murine models were used to examine the role of MBL in pdmH1N1 and H9N2/G1 virus infection. Results. Our study demonstrated that in vitro, MBL binds to pdmH1N1 and H9N2/G1 viruses, likely via the carbohydrate recognition domain of MBL. Wild-type mice developed more severe disease, as evidenced by a greater weight loss than MBL KO mice during influenza virus infection. Furthermore, MBL WT mice had enhanced production of proinflammatory cytokines and chemokines compared with MBL KO mice, suggesting that MBL could upregulate inflammatory responses that may potentially worsen pdmH1N1 and H9N2/G1 virus infections. Conclusions. Our study provided the first in vivo evidence that MBL may be a risk factor during pdmH1N1 and H9N2/G1 infection by upregulating proinflammatory response.

The pandemic H1N1 2009 (pdmH1N1) influenza A virus has spread globally since the outbreak first started in Mexico in 2009. As of August 2010, the virus had already caused more than 18 449 deaths in at least 214 countries worldwide [1] . The World Health Organization officially announced the step-down of pdmH1N1 to postpandemic phase on 10 August 2010, and the virus was expected to circulate as a seasonal virus in the human population thereafter [2] . Gene segments of the pdmH1N1 virus are derived from multiple lineages, including the Eurasian swine, the classical swine, and the triple reassortant swine lineages. Multiple genetic reassortment events of viral components have taken place and thus gave rise to this novel pandemic virus [3] . Clinical symptoms of pdmH1N1 infection are usually mild, possibly due to the cross-protection offered by memory cytotoxic T lymphocytes established from previous exposure to seasonal influenza [4] . H9N2 avian influenza virus (H9N2/G1) is widely prevalent among poultry in various Eurasian regions, including mainland China and Hong Kong. In 1999 and 2003, H9N2 influenza was reported in 3 children in Hong Kong. All 3 of them developed relatively mild symptoms and recovered within a week [5, 6] . The 6 internal genes of H9N2/G1 virus were found to be related to the highly pathogenic avian influenza virus A/Hong Kong/483/97 (H5N1). Although the H9N2/G1 virus was found mainly in poultry, it has the ability to transmit across species, and with its genetic similarity to the highly pathogenic H5N1, it also poses a threat of becoming pandemic [7] . Mannose-binding lectin (MBL) is a serum protein primarily produced by the liver. It belongs to the collectin family that comprises the collagen-like domain and the carbohydrate recognition domain (CRD). MBL functions as a key patternrecognition molecule recognizing a wide range of pathogens [8, 9] . Lectin pathway activation [10] and opsonophagocytosis are triggered upon MBL binding to pathogens [11] . While the MBL gene is highly polymorphic in humans, clinical association studies have demonstrated that MBL deficiency was associated with increased susceptibility to certain infections [12, 13] . The antiviral role of MBL in influenza virus infection remains controversial. Previous studies suggested that MBL demonstrates in vitro anti-influenza virus function, including inhibition of viral hemagglutination and direct neutralization of the virus either in a complement dependent or independent manner [14] [15] [16] . However, other studies have shown that the antiviral function of MBL may vary among different strains of influenza viruses, depending on the number of potential glycosylation sites on the viral hemagglutinin (HA) globular domain [17, 18] . Influenza virus-infected epithelial cells and macrophages can initiate a cellspecific response that includes the transcription and release of proinflammatory cytokines and chemokines [19, 20] . Although some studies have indicated that MBL may regulate proinflammatory cytokine and chemokine release from phagocytes in response to bacterial stimulation [21, 22] , little is known about its immunomodulatory role in influenza [23] . In the present study, we investigated whether MBL could display any in vitro or in vivo antiviral function toward pdmH1N1 and H9N2/G1 viruses, as well as whether it could modulate the inflammatory response upon infection by these two strains of influenza virus. Influenza virus A/California/04/2009 (pdmH1N1) were propagated in embryonated chicken eggs and purified by ultracentrifugation with minor modification of our previous work [4, 24] . Influenza virus A/Quail/Hong Kong/G1/97 (H9N2/G1) was grown in Madin-Darby canine kidney (MDCK) cells with modified Eagle's medium (Invitrogen) containing 2 lg/mL N-p-Tosyl-L-phenylalanine chloromethyl ketone (TPCK)treated trypsin (Sigma-Aldrich). Virus stocks were purified by adsorption to and elution from turkey red blood cells and stored at 280°C until use as previously described [25] . The determination of virus titer was performed by titrating virus in MDCK cells, with daily observation of cytopathic effect and confirmation by hemagglutination assay. The tissue culture infective dose affecting 50% of the cultures (TCID 50 ) was calculated by the Reed-Muench formula. Ultraviolet (UV)-irradiated virus was prepared by irradiation with energy of 0.2 J in a UV crosslinker as described previously [26] . The binding assay was performed as described previously [12] . In brief, 96-well flat-bottom polystyrene plates (Corning-Costar) were precoated with 100 lL/well of 10 2 , 10 3 ,10 4 , and 10 5 TCID 50 UV-irradiated influenza viruses or phosphate-buffered saline (PBS). After incubation at room temperature overnight, wells were blocked for 2 hours at room temperature with 1% bovine serum albumin (BSA) in PBS with 0.05% sodium azide. Different concentrations of recombinant human MBL (rhMBL) (0, 0.5, 2, 6, or 8 lg/mL), which was kindly provided by Dr K. Takahashi (Laboratory of Developmental Immunology, Harvard Department of Pediatrics, Massachusetts General Hospital, Boston), were added and incubated overnight at 4°C. Then 100 lL of 0.2 lg/mL biotinylated monoclonal anti-MBL antibody (HYB131-01, Antibody Shop) diluted in PBS with 1% BSA was added into each well. Bound antibody was detected by using horseradish peroxidase-conjugated streptavidin and tetramethybenzidine substrate solution (R&D Systems). The binding of MBL to influenza virus was evaluated by the absolute absorbance values measured at 450 nm (A 450 ). Breeding pairs of MBL wild-type (WT) and MBL knockout (KO) mice on C57B6/J were provided by Dr Takahashi [27] . They were maintained under specific pathogen-free conditions in the animal facilities of the Laboratory Animal Unit, The University of Hong Kong. Female mice were used at 6-10 weeks of age. They were anesthetized and inoculated intranasally with 30 lL of 10 3 TCID 50 pdmH1N1 virus, 10 5 TCID 50 H9N2/G1 virus, or PBS at day 0. Virus-infected or mocktreated mice were weighed daily. All animal care and experiments were conducted in accordance with the Committee on the Use of Live Animals in Teaching and Research guidelines of the University of Hong Kong. Virus-infected or mock-treated mice were sacrificed at days 3, 7, and 14 after infection. The lungs were harvested and homogenized by a tissue homogenizer (Omni International). The homogenates were centrifuged at 2500 rpm for 10 minutes at 4°C. Supernatants were used for virus titer determination and cytokine detection. Expression levels of interleukin (IL) 1a, IL-2, IL-4, IL-6, IL-10, tumor necrosis factor (TNF) a, interferon (IFN) c, macrophage inflammatory proteins (MIP)-1a, MIP-1b, monocyte chemotactic protein (MCP)-1, MCP-3, and Regulated upon Activation, Normal T-cell Expressed, and Secreted (RANTES) in the lung homogenates were quantitatively determined by flow cytometrybased immunoassay (Mouse Th1/Th2 cytokine 10plex and Mouse Chemokines 6plex Flowcytomix Multiplex, Bender MedSystems) according to the manufacturer's protocol. Interleukin 1b and keratinocyte chemoattractant (KC) simplex were purchased separately (Bender MedSystems). In brief, lung homogenates were prepared and processed accordingly. The samples were acquired on a BD LSRII (BD Bioscience), and the amount of cytokine (ng/mL) was calculated by FlowCytomix Pro 2.3 software (Bender MedSystems). Virus-infected or mock-treated mice were sacrificed at the indicated time point for histopathologic analysis. The lung tissues were fixed in 10% formalin and embedded in paraffin. Fivemicrometer-thick, paraffin-embedded sections were cut and stained with hematoxylin and eosin (H&E) to analyze histological lesions. Histopathologic score of lung tissues was examined by a board-certified pathologist blinded to the exposure status. Lung inflammatory changes were graded using a semiquantitative scoring system based on the following parameters: peribronchiolar and bronchial infiltrates, bronchiolar and bronchial luminal exudates, perivascular infiltrates, parenchymal pneumonia, and edema, as previously described [28] . Each parameter was graded on a scale of 0-4 with 0 as absent, 1 as slight, 2 as mild, 3 as moderate, and 4 as severe. The total lung inflammation score was expressed as the sum of the scores for each parameter. The degree of cell infiltration was independently scored on an increasing scale of 0-3 with 0 as no cells, 1 as few cells, 2 as moderate influx of cells, and 3 as extensive influx of cells [29] . Data were expressed as mean (standard error of the mean). Unpaired Student t test in GraphPad Prism 5.0 software (GraphPad) was used for statistical analysis. A P value ,.05 was considered significant. rhMBL Binds Both pdmH1N1 and H9N2/GI Viruses A microtiter capture assay demonstrated that MBL could bind to pdmH1N1 and H9N2/G1 in vitro ( Figure 1 ). The MBL-virus binding occurred in a dose-dependent manner ( Figure 1A and 1D). Similarly, increased amount of virus could also result in increased binding by MBL ( Figure 1B and 1E) . MBL utilizes the CRD to recognize pathogens in a calcium-dependent manner [30] . Further addition of ethylenediaminetetraacetic acid (EDTA) in the assay inhibited the binding of MBL to both strains of influenza virus ( Figure 1C and 1F) , suggesting that the binding occurred through the CRD of MBL. Wild-type and MBL KO mice, 6-10 weeks of age, were infected intranasally with 30 lL of 10 3 TCID 50 pdmH1N1 virus or 10 5 TCID 50 H9N2/G1 virus. The viral dosage chosen for the experiment was previously demonstrated to be sublethal (data not shown). Mice were inoculated with 30 lL of PBS as the mock treatment. No mice died throughout the 14-day experiment. The body weight of mice, which was a physiological value indicating infection progress and the health of the animals, was recorded daily. Mice receiving mock treatment did not lose body weight, demonstrating the absence of potential harmful effects due to anesthetics and intranasal inoculation (Figure 2A ). Both strains of influenza virus could successfully infect MBL WT and MBL KO mice as evidenced by the significant weight loss in these mice after virus infection. Compared to MBL KO mice, MBL WT mice had more weight loss upon virus infection. For pdmH1N1 virus infection, MBL WT mice showed a significantly greater body weight drop on days 3-12 compared with MBL KO mice ( Figure 2B ). The mean peak body weight loss observed in the MBL WT mice and MBL KO mice were -23.51% and -17.38%, respectively. For H9N2/G1 virus infection, MBL WT mice only showed a significantly greater weight loss on day 8 when compared with MBL KO mice ( Figure 2C ). Although similar mean peak weight loss was observed between the MBL WT mice and MBL KO mice after H9N2/G1 virus infection, MBL WT mice recovered more slowly than the KO mice. Collectively, these data suggested that the presence of MBL caused a more severe infection by the pdmH1N1 and H9N2/G1 viruses. Wild-type and MBL KO mice infected with pdmH1N1 or H9N2/G1 virus were sacrificed on days 3, 7, and 14 after infection. Virus titers in lung homogenates were determined by TCID 50 . As shown in Figure 3 , these 2 strains of influenza virus were detectable in the lung homogenates collected from MBL WT and KO mice on day 3 and day 7, confirming viral lung infection. On day 14, titers for both strains of virus were undetectable, which was consistent with the regain of body weight by MBL WT and MBL KO mice and suggested recovery from the infection. For pdmH1N1 virus infection, there was no significant difference in the lung virus titer between MBL KO and MBL WT mice on days 3 and 7 after infection ( Figure 3A ). In contrast, significantly less virus titer was detected in MBL KO mice on day 7 but not on day 3 after H9N2/G1 virus infection compared to that in MBL WT mice ( Figure 3B ). To further investigate whether MBL would modify the inflammatory response upon pdmH1N1 and H9N2/G1 virus infection, a panel of 14 proinflammatory cytokines and chemokines were examined in the lung homogenates collected from the infected MBL WT and MBL KO mice. The simultaneous profiling of the cytokines and chemokines was examined by using bead-based suspension array, which could allow the sensitive and specific detection of these proteins in the available amount of lung homogenates. As shown in Figures 4 and 5 , upon pdmH1N1 and H9N2/G1 virus infection, inflammatory response assayed by cytokines and chemokines production was triggered in both MBL WT and MBL KO mice. Except for IL-2, of which the level remained constantly low during the course of experiment, the kinetics of individual proteins were similar in that they were readily detectable on day 3, reached the highest level on day 7, and declined on day 14. Strikingly, MBL KO mice had reduced inflammatory responses during infection. Among the 14 cytokines examined, the majority of them showed significantly lower amounts in MBL KO mice lung homogenates than in MBL WT mice, including IL-1a, IL-1b, IL-6, IL-10, TNF-a, IFN-c ( Figures 4A and 5A) , KC, MIP-1a, MIP-1b, MCP-1, MCP-3 and RANTES (Figures 4B and 5B) . These results suggested that MBL upregulates the inflammatory response to influenza virus infection, resulting in elevated production of proinflammatory cytokines and chemokines in the MBL WT mice as compared with the MBL KO mice. To further confirm the severe inflammatory response in the MBL WT mice compared with MBL KO mice, lung sections were stained with H&E for histological analysis to evaluate inflammation-associated lung damage caused by pdmH1N1 and H9N2/G1 influenza virus infection. In the histological sections of MBL WT mice, more severe lung inflammation and more cell infiltration were observed when compared to that of MBL KO mice on day 7 ( Figure 6 ). Consistent with our cytokine and chemokine data, the pulmonary histological analysis suggested that the MBL WT mice had a more severe inflammatory response upon pdmH1N1 and H9N2/G1 virus infection. Mannose-binding lectin is a pattern-recognition molecule, which provides first line of host defense. Accumulating evidence has suggested that MBL exhibits in vitro anti-influenza virus properties by direct neutralization, inhibiting influenza virus hemagglutination, binding to the influenza virus as an opsonin, and activating the complement system through the lectin pathway [14, 16, 23] . However, these properties vary among different virus strains and subtypes [17, 18] . In this study, we focused on the pandemic influenza A H1N1 virus and avian influenza A H9N2/G1 virus, which are of potential threat to the global community. We demonstrated that despite these 2 strains of viruses being bound by rhMBL via the CRD of MBL at the physiological level, they infected both MBL WT and MBL KO mice effectively. Our results are consistent with a recent in vitro study by Job et al [18] , in which MBL was found to bind to pdmH1N1 fairly in vitro but the virus was resistant to the antiviral activity of MBL. The number and position of potential glycosylation sites on the viral HA globular domain determine the binding affinity between MBL and the virus. Even though MBL can physically bind to the virus, the binding may be insufficient for executing any antiviral function. Arguably, Chang et al [23] recently reported that MBL deficiency increases susceptibility to infection with influenza A virus Philippine 82 H3N2 (Phil82), which is a human strain. We reconcile with the suggestion that MBL effects would differ depending on strains of influenza A virus and thus MBL causes variable antiviral activities and host responses. The degree of glycosylation on the globular head of the HA molecule is believed to be essential for MBL to exhibit its antiviral properties. For Phil82 virus, the high-mannose oligosaccharide at residue 165 of the HA molecule has already been shown to be crucial for the neutralization by MBL [31] . Although pdmH1N1 virus contains a single potential glycosylation site at the base of the HA globular head (Asn 104 ), it lacks potential glycosylation sites on the globular head region of HA (Asn 142 , Asn 144 , Asn 172 , Asn 177 , and Asn 179 ) [18] . To our knowledge, binding of MBL on H9N2/G1 virus is not well documented in the literature. Therefore, we analyzed the potential glycosylation sites on HA of H9N2/G1 virus based on an in silico approach as suggested by Job et al [18] . The HA sequence data was retrieved from GenBank (AAF00706.1) and we used NetNGlyc 1.0 server to predict the number of potential glycosylation sites. We found that there was no potential glycosylation sites near residue 165 of the HA molecule of H9N2/G1 virus. We speculate that as a result of the absence of potential glycosylation sites near the receptor-binding domain of the HA globular head of both pdmH1N1 virus and H9N2/G1 virus, MBL fails to interfere with the viral binding to target cells despite its ability to bind the virus. This can adequately explain the discrepancy between the present in vivo data and Chang's study [23] . In this study, MBL WT mice were found to have a more severe disease in terms of greater weight loss and worse lung pathology than MBL KO mice during either pdmH1N1 or H9N2/G1 virus infection. This suggests that MBL may contribute to the disease severity seen in the MBL WT mice. To elucidate the mechanisms, we investigated the immune response, such as production of cytokines and chemokines at various time points during the infection. Most cytokines and chemokines were detected with similar kinetics, with the peak on day 7 following influenza virus infection in both MBL WT and MBL KO mice. These data suggest that the most critical phase of influenza infection occurs around day 7 after infection, and this is Interestingly, we found that MBL contributed to a more severe proinflammatory response by increasing the production of several proinflammatory cytokines, such as IL-1a, IL-1b, IL-6, TNF-a, and IFN-c. Interleukin 1a and IL-1b are multifunctional proinflammatory cytokines produced readily by influenzainfected leukocytes. They are capable of inducing fever, anorexia, and weight loss [32] . Enhanced production of these cytokines can contribute to the acute lung immunopathology after influenza virus infection in mice [33] and induce gene expression of other cytokines like IL-6 and TNF-a [34, 35] . Despite the abundance of IL-6 following the influenza virus infection, in vivo studies showed that it does not contribute significantly to the pathogenesis of influenza virus infection because the mortality and morbidity observed in mice infected with H5N1 are comparable in both MBL WT and IL-6 deficient mice [36, 37] . Tumor necrosis factor a is readily produced by influenza virusinfected leukocytes and can activate macrophages, stimulate dendritic cell maturation and neutrophils, further enhance the inflammatory response, and activate efficient antigen presentation system in the infected site [20, 38] . Excessive production of TNF-a causes tissue injury, hemorrhagic shock, and death in mice [39, 40] . Interferon c is also an important proinflammatory cytokine that has different functions, including the activation of macrophages, differentiation of Th1 from T cells, enhancement of antigen presentation, and expression of the chemokine gene [41, 42] . These proinflammatory cytokines are commonly found in the acute-phase response to influenza virus infection and may induce immunity but also cause damage to the host tissue [43] . These cytokines were also increased in our infected murine lung, with significantly higher levels in MBL WT mice than in MBL KO mice on day 7, coinciding with body weight loss and lung histological findings. Interleukin 10, an antiinflammatory cytokine, was also significantly higher in MBL WT mice than in MBL KO mice on day 7. Interleukin 10 deficiency was reportedly protective in high-dose influenza virus infection [44] , implying that increased IL-10 may be deleterious to the host. As a consequence of such an overwhelming ''cytokine storm'' [45, 46] , the MBL WT mice were found to have a worse disease course than in the MBL KO mice, including greater body weight loss and more severe lung inflammation. In addition, we also found that most chemokines, including KC, MIP-1a, MIP-1b, MCP-1, and MCP-3, were elevated in MBL WT mice compared with the MBL KO mice in both pdmH1N1 and G1/97 virus infection. Influenza virus-infected macrophages produced large amounts of these chemokines in vitro [25, 47, 48] . Functionally, these chemokines are important mediators for immune cell activation and chemotactic factors, which recruit leukocytes to the infected sites [49] . This may help account for our histological observation that more inflammatory cell infiltration was observed in MBL WT mice than in MBL KO mice. The role of MBL in modulating immune responses has also been observed in Staphylococcus aureus infection. It was shown that MBL amplifies the host immune response during S. aureus infection by cooperating with Toll-like receptors 2 and 6 and augments the production of proinflammatory cytokines and chemokines [50] . The observation from our present study prompted us to further investigate whether MBL may also cooperate with other pattern-recognition receptors and thus further amplify the host response during influenza virus infection. In conclusion, we have shown for the first time that MBL is a risk factor leading to a more severe pdmH1N1 and H9N2/G1 virus infection by upregulating proinflammatory responses. Financial support. This work was supported in part by the National

Essential epidemiological mechanisms underpinning the transmission dynamics of seasonal influenza

Seasonal influenza has considerable impact around the world, both economically and in mortality among risk groups, but there is considerable uncertainty as to the essential mechanisms and their parametrization. In this paper, we identify a number of characteristic features of influenza incidence time series in temperate regions, including ranges of annual attack rates and outbreak durations. By constraining the output of simple models to match these characteristic features, we investigate the role played by population heterogeneity, multiple strains, cross-immunity and the rate of strain evolution in the generation of incidence time series. Results indicate that an age-structured model with non-random mixing and co-circulating strains are both required to match observed time-series data. Our work gives estimates of the seasonal peak basic reproduction number, R(0), in the range 1.6–3. Estimates of R(0) are strongly correlated with the timescale for waning of immunity to current circulating seasonal influenza strain, which we estimate is between 3 and 8 years. Seasonal variation in transmissibility is largely confined to 15–30% of its mean value. While population heterogeneity and cross-immunity are required mechanisms, the degree of heterogeneity and cross-immunity is not tightly constrained. We discuss our findings in the context of other work fitting to seasonal influenza data.

Seasonal influenza causes significant levels of morbidity and mortality around the world each year, yet its dynamics and the annual sequence of pathogen subtypes are hard to predict [1] . Understanding the mechanisms underlying the annual behaviour of influenza and their sensitivity to parameters is important for the forecasting and control of seasonal epidemics and also in assessing the effect of the introduction of novel antigenic strains. The 2009 H1N1 pandemic is a recent example [2] . The initial outbreak of this novel H1N1 strain occurred in spring and summer, 'out of season' in the Northern Hemisphere. Uncertainty with regard to intensity of transmission during summer substantially complicated forecasting of the likely trajectory of the epidemic over the following months. Many respiratory transmissible diseases exhibit seasonal epidemics. Annual periodic forcing causes a wide range of oscillatory epidemic behaviour, as illustrated by incidence rates for measles and pertussis [3] [4] [5] . The source of seasonal forcing in those cases is largely attributed to annual variations in the intensity of contact between children, but a range of other mechanisms have been proposed. In the case of influenza, two recent hypotheses have centred on variation in vitamin D levels and air humidity [6, 7] . Influenza dynamics are additionally complicated by considerable antigenic diversity in human influenza viruses. Within each of the two influenza A subtypes (H3N2 and H1N1) co-circulating prior to 2009, continual evolution selecting for antigenic novelty was seen [8] . The dynamics of intra-subtype evolution is characterized by relatively stable strains periodically replaced by antigenically distinct types in punctuated evolution events [8] [9] [10] . In addition, there is evidence of immune-mediated competition between subtypes [10] . These mechanisms combine to generate complex seasonal behaviour, featuring a range of annual attack rates (AARs) and epidemic durations and alternating sequences of dominant annual subtypes and strains [11] . Detailed and computationally intensive simulations are required to fully integrate the epidemiological and genetic aspects of long-term behaviour [10, 12] . However, strain-specific data with sufficient temporal resolution to fit such a model are not available. We therefore adopt a simplified description here and model two weakly interacting subtypes, with evolution within each subtype being represented by a gradual loss of immunity of the previously exposed host population, a so-called SIRS (susceptible-infected-recoveredsusceptible) model. While use of SIRS models to represent intra-subtype evolution is not uncommon [13] [14] [15] , previous models of influenza time series have not accounted for dynamics generated by multiple interacting subtypes. In this paper, we identify a set of key features characterizing seasonal influenza-like illness (ILI) incidence time series and use these to define information measures for the distance between a transmission model and empirical observations. Use of summary statistics (rather than attempting to directly fit models to timeseries data) adds robustness to variations in reporting (much ILI is not even caused by influenza) and shortterm fluctuation in the incidence data. Our aim is not to precisely match the incidence time series, but to find broad regions of parameter space which are consistent with the characteristics of seasonal flu epidemics and hence identify which epidemiological mechanisms are essential and acceptable ranges for their parameters. For this purpose, an elaborate model and detailed data are not necessary. We use the following features of the time series to characterize seasonal influenza incidence in temperate countries of the Northern Hemisphere: -Epidemic duration. The vast majority of seasonal influenza outbreaks are observed between the extremes of mid-November and the end of April [11] . Since a background rate of ILI incidence is present throughout the year, duration is difficult to define precisely. The range of values quoted for the duration of an annual epidemic (described with the acronym ADE in this paper) is typically between three and 16 weeks [1, 16] . -Attack rate. The proportion of the population infected in a year, which we term the AAR, is very difficult to determine, as a significant proportion of infections are asymptomatic and only a proportion of symptomatic cases seek healthcare. Hence sentinel data on ILI can only give relative information, such as the fractional variation in incidence over time. In addition, measures such as ILI are non-specific for influenza and can be caused by a variety of respiratory pathogens. That said, ILI data from France and the UK indicate a standard deviation for AAR across successive years of approximately 40 per cent of the mean attack rate [17, 18] . Results from serological and virus isolation data from closely monitored populations indicate an AAR range of approximately 10-20% for seasonal influenza, rising to 30 per cent or more for influenza pandemics [1, 17] . Information on consultation rates for known cases also suggests a mean AAR of around 15 per cent [13, 18, 19 ]. -Periodic behaviour. Seasonal influenza is characterized by annual outbreaks in temperate countries in the sense that there is almost invariably a marked seasonal increase in case rate during the winter months. However, it cannot be said to be periodic in the sense that successive outbreaks are comparable in magnitude and form with each other. In this sense, seasonal influenza has no clear periodicity (annual, biennial and triennial) and contrasts with the behaviour of diseases such as measles, which has a pronounced biennial structure in the pre-immunization period [5] . The metrics that we use to compare model and time-series behaviour are able to pick up this characteristic aperiodicity. -Strain variation. Virus isolation studies show that individual seasonal epidemics are usually dominated by a single type and/or subtype, although others may be present at low levels [13] . Successive seasons are usually dominated by different types and subtypes, although often the same strain is present for several seasons (figure 1). Seasonal influenza is characterized by annual outbreaks in temperate countries in the sense that there is almost invariably a marked seasonal increase in case rate during the winter months. However, it cannot be said to be strictly periodic in the sense that successive outbreaks are comparable with each other in magnitude and duration. In this sense, seasonal influenza has no clear periodicity and contrasts with the behaviour of diseases such as measles, which has a pronounced biennial structure in pre-immunization period [5] . We use a deterministic SIRS compartmental model, to which we have added a number of refinements to capture critical aspects of influenza infection and immunity. The basic dynamics of infection and immunity in the model are where l j is the force of infection of strain j and S n is the part of the population that is immune to strain n. Infected individuals recovering from strain n enter a class entirely immune to n (e.g. S 0 ! S 1 , S 1 ! S 12 ). Immunity to a given strain is lost at a rate s. Figure 2 illustrates the flows of individuals between the various immunity classes. The model is further complicated by stratification by age (children and adults) with heterogeneous mixing, differential infectiousness and susceptibility by age and a realistic infectiousness profile. The full details of the model are described in the electronic supplementary material. The effect of seasonal forcing is included through a time-varying contact parameter, Mechanisms of seasonal influenza J. Truscott et al. 305 We express this variation in terms of peak R 0 and relative amplitude of variation throughout. The sinusoidal form is a good match for variation in infectiousness as a function of absolute humidity in temperate regions [14] , but we also examine the effect of using school terms as a seasonal driver using step-function forcing (see the electronic supplementary material). In determining feasible ranges for these parameters, we reviewed the range of estimates that exist for R 0 . The majority of these are calculated for the major global epidemics (1918, 1957, 1968) , since the antigenic novelty of pandemic viruses means an assumption of a serologically naive population can be made, making analysis simpler. For seasonal influenza, knowledge of population susceptibility is necessary to estimate R 0 (as opposed to the effective reproduction number, R). Estimated values for the 1918 pandemic range from 1.3 to 2.8 [20] . Similar values are found for the 1957 and 1968 epidemics [21] . These values also correspond well with those from studies of seasonal influenza, giving winter R 0 of 1.7 and school holiday R 0 of 1.4 [16] . The periodic forcing of systems of ordinary differential equations leads to a rich variety of behaviour, characterized by solutions with periods that are multiples of the forcing period (see [22] for SEIR example). Typically, smooth variation of the parameters can lead to sudden changes in the periodicity of the stable behaviour of the system. As will be seen in §3, the goodness of fit of the model is strongly dependent on the periodicity of the model's solution. We represent the generation time for the pathogen by an Erlang distribution with shape parameter k ¼ 4 and mean 1/a, where a ¼ 2.7 days [20, 23] . Our model incorporates two strains of influenza, for instance, representing H1N1 and H3N2, to try and capture aspects of the co-circulation of multiple influenza types and subtypes [11] . There are four immune states for individuals in the model; entirely susceptible, immune to either strain 1 or 2 and immune to both strains. The formulation allows for the inclusion of a basic cross-immunity mechanism, whereby an individual infected with either strain has a probability, f, of becoming immune to the other as well (assuming that this was not already the case) The flow between different immune states is illustrated in the electronic supplementary material. We note that this cross-immunity response is different from the shortterm non-specific response with regard to influenza strains considered elsewhere [10] , though cross-immunity is assumed to wane at the same rate as strain-specific immunity. Our model also includes a mechanism for loss of immunity, returning individuals to a susceptible state. Surveillance data show that human influenza strains can be grouped into clusters within each of which there is a high level of cross-immunity, but between which cross-immunity is much lower [8] . The appearance of a new cluster therefore corresponds to a step change in the susceptibility of the population to the current strain. Our model caricatures this process with a timescale, D, for resistant individuals to become susceptible to the current strain again. As antigenically distinct clusters appear every 2 -8 years [8, 24] , this is our expected range for values of D. There is evidence from contact studies and from modelling of influenza epidemics that infectious contact between individuals is highly assortative and agedependent, with the highest rates among school-age children [16, 25, 26] . We include these effects by stratifying the population into children (less than or equal to 14 years) and adults (more than 14 years) and employing a mixing matrix to describe contact between the two groups. The degree of assortativity is controlled by the parameter, u, and can be varied between random mixing (u ¼ 0), where groups contact each other proportional to the fraction of the population they represent, and wholly assortative (u ¼ 1) where each group mixes only with itself. Differences in intensity of contact are captured by relative susceptibility and infectiousness parameters, r and c (see the electronic supplementary material for details). To assess the quality of fit of the model behaviour to the data, we compare the distribution of key features in the time-series data with those generated by the epidemic model using the Kullback-Leibler (KL) information distance. We use normal distributions to characterize the empirical distributions of AAR and epidemic duration across a number of years. As discussed above, ignorance of the reporting rate makes it hard to know the underlying 'real' infection rate and also makes it difficult to compare reported incidence collected under different surveillance systems. In order to compare the data from the UK and France, we assume constant reporting rates for the UK and French surveillance systems, respectively, and scale the reported values linearly such that each has a mean AAR of 15 per cent (see the electronic supplementary material). Both dataset yield standard deviations of around 35 per cent of mean value for AAR and 11 + 2 weeks for epidemic duration. We calculate the KL information distance between model and data, I, for each of the key features as follows: where f is the distribution taken from the data, g is the approximate distribution of the same feature recovered from the model over many simulated years and p is a vector of model parameters. (See the electronic supplementary material for implementation.) The overall measure of goodness of fit used is the unweighted sum of the information distances for AAR and duration. We explore parameter space to identify regions where model behaviour most closely resembles empirical patterns. Although a simplified description of the epidemiological and evolutionary mechanisms of human influenza, our model nevertheless incorporates a substantial number of parameters. We focus on the following groupings: -the seasonal peak value of R 0 , here termed R p , and its relative amplitude d (R p (1 -d) being the seasonal minimum value of R 0 ). Strictly, these parameters control overall transmissibility and the magnitude of seasonal forcing of transmission; -the timescale for the generation of antigenically new strains, represented by mean duration of immunity to the current influenza strain, D, and the crossimmunity between strains, f ( §2); -the degree of assortativity in the contact patterns between children and adults, u; -external force of infection, e, representing effect of contact between members of the modelled population and infected individuals outside the modelled population. The values of other parameters are listed in table 1 and discussed in the electronic supplementary material. In discussing the behaviour of the model, periodicity refers to the periodicity of the overall case rate with time, rather than for an individual strain. Simulations were run using a population of 60 million, approximating the population of the UK. Owing to the large population, a deterministic model was used. Tests using the corresponding stochastic models showed no qualitative variation from the deterministic dynamics and the presence of a continuous low level external force of infection precluded the possibility of extinction. Figure 3 illustrates the behaviour of the model and aspects of the information distance between model and data as a function of R p and D. The region of best fits is located in a narrow diagonal band spanning 1.6 , R p , 2.5 and 3 , D , 8 (figure 3d ). Along this band, increasing reproduction number is compensated for by a longer period of effective immunity that decreases the susceptible proportion of the population, giving a constant mean attack rate. The acceptable region is bounded in part by the duration of the model epidemics. The lowest and highest values of R p generate epidemics that are too broad and too narrow, compared with the target distribution. The fit of the model is strongly constrained by the dynamics of the model, which exhibits a wide range of periodic behaviour across quite small changes in parameter values (figure 3a). While AAR changes smoothly with the parameters, abrupt changes in periodicity lead to qualitative changes in the distributions of AAR and epidemic duration and hence the KL distance. Best-fit behaviour is associated with long-period behaviour of the model (4þ years). Here, the model generates a range of AARs clustered around the mean and matching the target distribution. The qualitatively different forms of behaviour are well characterized by the total KL distance measure (figure 4). KL distances less than 200 correspond to realistic behaviour with appropriate mean attack rate distributions (figure 4a). KL distance between 200 and 300 match either with realistic behaviour interspersed with large-scale epidemic episodes or with realistic behaviour but with a mean attack rate displaced from the target value ( figure 4b) . Larger values represent dynamics and mean attack rates greatly different from the observed time series (figure 4c). Figure 5 illustrates a strong sensitivity to the amplitude of variation of the contact parameter, d, with the best-fit lying in the range 0.15-0.3. Although the mean AAR is not strongly dependent on d (figure 5a), the bifurcation behaviour of the model means realistic solutions (resembling figure 4a,b) can be found for higher amplitudes of seasonal variation but not in a contiguous region (figure 5b). Solutions with smaller seasonal variations are rejected on the epidemic duration component of the information distance. Low amplitudes generate broad epidemics which do not match the target distribution. The behaviour of our model is quite sensitive to the assumed external force of infection, e. The level of external forcing assumed for most of this work ( §2) is negligible compared with the average force of infection generated by the indigenous population. However, long-period and chaotic solutions for seasonally forced SIR models generate very low infection prevalence during epidemic troughs, even low levels of importation of infectives strongly encourages annual and biennial behaviours and removes highly chaotic solutions from the optimal region. The effect of importation rate can be seen in figure 6 . For external forces of infection above about 10 25 yr -1 , only annual and biennial solutions are found. Optimal behaviour is found for an external force of infection of approximately 6 Â 10 27 yr -1 . Figure 7 explores the sensitivity of the model to the degree of heterogeneity and cross-immunity in the population. The choice of R p and D lies in the wellfitting band in figure 3d. It is clear that a wide range of values for these parameters allow the model to fit the behaviour of the time series quite well. The model's qualitative behaviour (in terms of its periodicity) is more stable with respect to these mechanisms and variation within the closest fit region mainly affects the mean attack rate. There is a broadly inverse relationship between the well-fitting values of the parameters u and f. Increasing the assortativity of mixing concentrates infections more strongly in age groups, decreasing the available susceptibles and hence the attack rate. Increasing cross-immunity has an equivalent effect by increasing the effective loss of susceptibles caused by any single infection event and hence reducing the attack rate. As can be seen from figure 7a, values of the cross-immunity parameter f outside the range 0.3-0.6 drive the model into unfavourable periodicities, giving very poor fits. This suggests that a model with two strains interacting via cross-immunity is necessary to reproduce the dynamics seen in influenza time series and that it is insufficient to have two independent strains (f ¼ 0) or two antigenically identical strains (i.e. f ¼ 12equivalent to a single strain model). Similarly, extreme values of u also lead to poorly fitting model behaviour, suggesting that a uniformly mixing population (u ¼ 0) would also not generate matching behaviour. In this work, we have identified a minimal set of mechanisms necessary to match the long-term temporal behaviour seen in ILI time series. We find that an age-structured population and multiple strains with cross-immunity are necessary to recreate the distributions of AAR and epidemic duration seen in time-series data. Nonlinear models of this type with temporal forcing are well known for having complex bifurcation structures affecting their periodic behaviour. On the timescale of a single disease season or single epidemic, these would have little effect on parameter estimation. Over many seasons, however, the periodicity of the underlying model is crucially important. We would argue that capturing the long-term trends in behaviour is at least as important as the detail of individual seasonal outbreaks and our fitting approach focuses on these aspects. The complex bifurcation structure of the model means that the information distance is not necessarily a smooth function of the parameters. This makes finding a unique set of parameters giving an overall minimum distance impossible. Although mean attack rate and duration generally vary smoothly, the period of the model solution changes discontinuously (inevitably, as it only takes values that are multiples of the annual forcing period). Solutions with longer periods generate a wider range of AAR and epidemic durations over an extended period of time and are therefore capable of fitting the target data distributions better. As a result, the best fits are strongly associated with longer periods in model solutions and the quality of fit of the model can change abruptly over small changes in parameter values. Peak R 0 , the amplitude of variation of R 0 and the duration of immunity are all strongly constrained. Figure 3 illustrates that increasing R 0 is offset by a longer duration of immunity reducing the susceptible population. The narrowness of the well-fitting parameter region is a result of the sudden changes of behaviour generated by changes in these parameters. Figure 5 also illustrates this feature. The closest fitting region is found for d between 0.15 and 0.3, but patches of well-fitting solutions are scattered a range of values of R p and d owing to the sensitivity to the system to temporal forcing. We note that a change in the mode of forcing from sinusoidal to school-term leads to generally broader ranges of acceptable parameter values (see the electronic supplementary material), perhaps indicating that the presence of this mechanism is a strong contributor to the variable annual behaviour observed in the ILI dataset. Because of the difficulties in knowing the 'true' incidence rate, the mean AAR is not precisely known and a range of 10-20% is often quoted. To allow for this uncertainty, we investigated allowing the information distance calculation to be based on the best-fit mean AAR from the range 10-20%, rather than precisely 15 per cent per year. Resulting best-fit parameter regions for R 0 against D and u against f were not significantly changed, owing almost certainly to the dominance of qualitative model behaviour as described above. Incidence periodicity of the model is much less sensitive to cross-immunity and age structure, resulting in a wide region of close fitting behaviour for values of u between about 0.2 and 0.6 and f between 0.3 and 0.5. As discussed in §3, this strongly suggests that an agestructured population and, in particular, a pathogen population with more than one strain and crossimmunity, are necessary to reproduce the patterns of behaviour found in the ILI time series. Models based on single strains and well-mixed populations generate annual and biennial behaviour for the same parameter values. The necessity of multiple strains has been noted in other work modelling seasonal influenza [14] , although in that work the two strains did not interact. Within the best-fitting parameter regimes, incidence periodicity for each modelled strain is basically biennial with strains dominating alternate years. While real strain dynamics are clearly more complex than this (figure 1), observed patterns do show a tendency for a particular strain not to dominate in successive years. Within the model, cross-immunity generates a negative correlation between strains, causing them to alternate in successive years. For strong cross-immunity, both strains become antigenically similar and goodness of fit falls off rapidly. It is instructive to compare our results with those of previous papers fitting simple models to influenza incidence data. Work by Xia et al. [15] used a simple single strain SIRS, but with a more detailed description of temporal variation in contact rate and loss of immunity post-recovery. In addition, the infection rate was described by the phenomenological term b I a g(S). Exponents of this type are well known to facilitate fitting [5] , but are hard to interpret. We note that both the form of this term and the non-exponentially distributed duration of immunity in that study affect epidemic attack rates as a function of transmissibility and the periodicity of epidemics, and hence may play an equivalent role to age structure and the incorporation of two subtypes in our model in allowing a good fit to the data. Shaman et al. [14] use a similar SIRS model to investigate the possibility that seasonal changes in absolute humidity can generate recorded patterns in pneumonia and influenza mortality data. The model was stochastic, has no age structure, and effectively uses only one strain. Best-fit parameter values are similar to those found in this work, although the relative amplitude of variation in R 0 is in the range 0.4-0.5, significantly higher than our findings. Best-fit parameter sets showed considerable lack of correlation with each other, which the authors attribute to the stochastic nature of their model. Our work suggests that this scatter may be the result of the complex bifurcational structure of such models. As already discussed, our results indicate that a two strain model without cross-immunity or age structure is unlikely to fit patterns of seasonal influenza from a temperate region. However, there are several significant differences between the two systems. Shaman et al. employ a significantly higher background force of infection than ours (approx. 7.3 Â 10 24 yr -1 ), which would place our model in a strongly annual or biennial regime. Hence that model may reproduce the mean attack rate well, but not its variability. Our assessment of goodness of fit is currently focused primarily on distribution of AARs and duration of epidemics, although we also take account of the sequence of strains and the age-distribution of cases (see the electronic supplementary material). Future work will test our conclusions against a full description of the incidence data as well as against different choices of key features in the data, such as time of epidemic onset.

Amiodarone Exposure During Modest Inflammation Induces Idiosyncrasy-like Liver Injury in Rats: Role of Tumor Necrosis Factor-alpha

Amiodarone [2-butyl-3-(3′,5′-diiodo-4’α-diethylaminoethoxybenzoyl)-benzofuran] (AMD), a class III antiarrhythmic drug, is known to cause idiosyncratic hepatotoxic reactions in human patients. One hypothesis for the etiology of idiosyncratic adverse drug reactions is that a concurrent inflammatory stress results in decreased threshold for drug toxicity. To explore this hypothesis in an animal model, male Sprague-Dawley rats were treated with nonhepatotoxic doses of AMD or its vehicle and with saline vehicle or lipopolysaccharide (LPS) to induce low-level inflammation. Elevated alanine aminotransferase (ALT), aspartate aminotransferase, alkaline phosphatase, and gamma-glutamyltransferase activities as well as increased total bile acid concentrations in serum and midzonal hepatocellular necrosis were observed only in AMD/LPS-cotreated rats. The time interval between AMD and LPS administration was critical: AMD injected 16 h before LPS led to liver injury, whereas AMD injected 2–12 h before LPS failed to cause this response. The increase in ALT activity in AMD/LPS cotreatment showed a clear dose-response relationship with AMD as well as LPS. The metabolism and hepatic accumulation of AMD were not affected by LPS coexposure. Serum concentration of tumor necrosis factor-alpha (TNF) was significantly increased by LPS and was slightly prolonged by AMD. In Hepac1c7 cells, addition of TNF potentiated the cytotoxicity of both AMD and its primary metabolite, mono-N-desethylamiodarone. In vivo inhibition of TNF signaling by etanercept attenuated the AMD/LPS-induced liver injury in rats. In summary, AMD treatment during modest inflammation induced severe hepatotoxicity in rats, and TNF contributed to the induction of liver injury in this animal model of idiosyncratic AMD-induced liver injury.

Amiodarone [2-butyl-3-(3#,5#-diiodo-4'a-diethylaminoethoxybenzoyl)-benzofuran] (AMD), a class III antiarrhythmic drug, is known to cause idiosyncratic hepatotoxic reactions in human patients. One hypothesis for the etiology of idiosyncratic adverse drug reactions is that a concurrent inflammatory stress results in decreased threshold for drug toxicity. To explore this hypothesis in an animal model, male Sprague-Dawley rats were treated with nonhepatotoxic doses of AMD or its vehicle and with saline vehicle or lipopolysaccharide (LPS) to induce low-level inflammation. Elevated alanine aminotransferase (ALT), aspartate aminotransferase, alkaline phosphatase, and gamma-glutamyltransferase activities as well as increased total bile acid concentrations in serum and midzonal hepatocellular necrosis were observed only in AMD/LPS-cotreated rats. The time interval between AMD and LPS administration was critical: AMD injected 16 h before LPS led to liver injury, whereas AMD injected 2-12 h before LPS failed to cause this response. The increase in ALT activity in AMD/LPS cotreatment showed a clear doseresponse relationship with AMD as well as LPS. The metabolism and hepatic accumulation of AMD were not affected by LPS coexposure. Serum concentration of tumor necrosis factor-alpha (TNF) was significantly increased by LPS and was slightly prolonged by AMD. In Hepac1c7 cells, addition of TNF potentiated the cytotoxicity of both AMD and its primary metabolite, mono-N-desethylamiodarone. In vivo inhibition of TNF signaling by etanercept attenuated the AMD/LPS-induced liver injury in rats. In summary, AMD treatment during modest inflammation induced severe hepatotoxicity in rats, and TNF contributed to the induction of liver injury in this animal model of idiosyncratic AMD-induced liver injury. Key Words: amiodarone hepatotoxicity; inflammation; lipopolysaccharide; idiosyncratic adverse drug reactions; drug metabolism; tumor necrosis factor-alpha. Idiosyncratic adverse drug reactions (IADRs) typically occur only in a small fraction of patients who are treated with certain drugs at therapeutic doses. IADRs are usually unrelated to the pharmacological target of the drug. They present a serious human health problem and are usually not predicted by current preclinical safety evaluation during drug development. During the period 1975-2000, 10% of newly approved drugs were withdrawn from the U.S. market or received black box warnings due to these adverse reactions Uetrecht, 2007) . The mechanisms by which IADRs occur are not clear. Evidence from experimental animals indicates that mild inflammation can decrease the threshold for toxicity and thereby render an individual susceptible to an adverse drug reaction that would not otherwise occur . Lipopolysaccharide (LPS), a cell wall component of gram-negative bacteria, is widely used as an inflammagen in these animal studies. A nonhepatotoxic dose of LPS can interact with nontoxic doses of several IADR-associated drugs from different pharmacologic classes to induce liver damage in rodents (Deng et al., 2006; Luyendyk et al., 2003; Waring et al., 2006; Zou et al., 2009b) . Amiodarone [2-butyl-3-(3#,5#-diiodo-4'a-diethylaminoethoxybenzoyl)-benzofuran] (AMD), a class III antiarrhythmic drug, is effective in increasing the survival of patients after myocardial infarction or congestive heart failure (Singh, 1996) . Since the approval of AMD by the U.S. Food and Drug Administration in 1985, the use of this drug has been associated with a variety of adverse effects, including liver dysfunction, pulmonary complications, thyroid dysfunctions, and ocular disturbance (Rotmensch et al., 1984) . The reported frequency of liver abnormalities in patients receiving AMD varies from 14 to 82% (Lewis et al., 1989) . Most of these reactions are mild, with serum transaminase elevation within threefold of the upper limit of normal (ULN), but some are more severe (Babatin et al., 2008) . Cases of liver reactions after intravenous administration of AMD are rare, but damage can be acute and marked (Rätz Bravo et al., 2005) . Fulminant hepatic failure or death associated with AMD hepatotoxicity has also been reported (Babatin et al., 2008) . There is evidence that the interaction between LPS-induced cytokines and drugs or their metabolites plays an important role in the LPS-drug interaction (Zou et al., 2009a) . Tumor necrosis factor-alpha (TNF) is a proximal mediator of the inflammatory cascade induced by LPS (Beutler and Kruys, 1995) and is critically involved in many models of liver injury, such as ischemia/reperfusion (Teoh et al., 2004) , alcoholic liver disease (Yin et al., 1999) and some drug/LPS-induced liver injury models (Shaw et al., 2009b; Tukov et al., 2007; Zou et al., 2009a) . As an example, TNF selectively augmented the cytotoxicity of sulindac sulfide, which is the major toxic metabolite of sulindac (Zou et al., 2009a) . In the case of amiodarone, the major metabolite of AMD is mono-N-desethylamiodarone (DEA), which shares similar pharmacological (Talajic et al., 1987) and pharmacokinetic (Shayeganpour et al., 2008) characteristics with AMD. DEA has antiarrhythmic properties, a very long half-life, and accumulates in the liver and many other tissues. In primary hepatocytes, HepG2 cells and other cell types, DEA is much more cytotoxic than AMD (Waldhauser et al., 2006) . The plasma concentration of DEA is greater in cases of AMD-associated IADRs (O'Sullivan et al., 1995) , suggesting a possible role for this metabolite in AMD toxicity. The purpose of this study was to test the hypothesis that inflammatory stress induced by LPS potentiates amiodaroneinduced hepatotoxicity in rats. When the results demonstrated a hepatotoxic interaction between inflammatory stress and amiodarone, the roles of metabolism and TNF were explored. Materials. Unless otherwise noted, all chemicals were purchased from Sigma-Aldrich (St Louis, MO). The activity of LPS (Lot 075K4038, derived from Escherichia coli serotype O55:B5) was 3.3 3 10 6 endotoxin units (EU)/ mg, which was determined by a Limulus Amebocyte Lysate Kinetic-QCL kit from Cambrex Corp. (Kit 50-650U; East Rutherford, NJ). The reagents for the measurement of alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), and gamma-glutamyltransferase (GGT) activities were purchased from Thermo Electron Corp. (Waltham, MA). The kit for total bile acids measurement was purchased from Diazyme Laboratories (Poway, CA). Animals. Male Sprague-Dawley rats (Crl:CD(SD)IGS BR; Charles River, Portage, MI) weighing 250-370 g were used for in vivo studies. They were fed standard chow (Rodent Chow/Tek 8640; Harlan Teklad, Madison, WI) and allowed access to water ad libitum. Animals were allowed to acclimate for 1 week in a 12-h light/dark cycle prior to experiments. They received humane care according to the criteria in the Guide for the Care and Use of Laboratory Animals. Experimental protocol. In all the experiments, rats were fasted for 12 h before administration of LPS and food was returned thereafter. A 20 mg/ml solution of AMD was made in its vehicle (0.18% Tween 80), and 4.1 3 10 5 EU/ml solution of LPS was made in sterile saline. To determine the optimal time interval between AMD and LPS treatments, rats were treated with AMD (300 mg/kg, ip) 2 h, 8 h, 12 h, 16 h, or 20 h before LPS (1.6 3 10 6 EU/kg, iv). For the evaluation of the dose-response relationship for AMD, rats were treated with AMD (0-400 mg/kg, ip) and 16 h later with LPS (1.6 3 10 6 EU/kg, iv) or saline. For the evaluation of the dose-response relationship for LPS, rats were treated with AMD (400 mg/kg, ip) or vehicle and 16 h later with LPS (0-1.6 3 10 6 EU/kg, iv). In subsequent studies, rats were treated with AMD (400 mg/kg, ip) or vehicle and 16 h later with LPS (1.6 3 10 6 EU/kg, iv) or saline. In the etanercept treatment study, rats were treated with etanercept (8 mg/kg) or sterile water by sc injection 1 h before LPS. Rats were anesthetized with isoflurane, and blood and liver samples were taken. Serum was prepared from blood, and plasma was prepared from blood collected into a syringe containing 3.2% sodium citrate (BD Biosciences, San Diego, CA). The right medial lobe of the liver was rapidly frozen for immunohistochemistry, and the left lateral lobe of liver was fixed in 10% neutral-buffered formalin and stored in 70% ethanol for histopathology. Evaluation of liver injury. Liver injury was estimated from the serum activities of ALT, AST, ALP, and GGT and from the serum concentration of total bile acids. Formalin-fixed liver samples were embedded in paraffin, sectioned, and stained with hematoxylin and eosin (H&E) staining. The stained liver sections were examined using light microscopy. Drug and metabolite analysis. Serum samples and liver homogenates were mixed with acetonitrile containing ethopropazine as internal standard (IS). After vortexing and centrifugation, protein was removed, and the supernatant was diluted and transferred to autosampler vials for liquid chromatographytandem mass spectrometry (LC/MS/MS) analysis. LC/MS/MS analysis was performed by use of a Shimadzu LC-20 high performance liquid chromatography (HPLC) system coupled to a QTRAP 3200 tandem quadrupole mass spectrometer (AB SCIEX, Foster City, CA) operated under control of Analyst v. 1.4.2 software. The Ascentis Express C18 HPLC Column (5 cm 3 2.1 mm, 2.7 lm) was maintained at 50°C. A volume of 2 ll was injected into the HPLC system and eluted with a gradient based on 10mM ammonium acetate in H 2 O (solvent A) and methanol (solvent B): 0-0.5 min, 10% solvent B; 0.5 -1 min, 10-98% solvent B; 1-4 min, 98% solvent B; 4-6 min, 10% solvent B; flow rate, 0.3 ml/min. Positive mode electrospray ionization was used for all analyses. Mass spectrometry parameters, including declustering potential and collision energy, were optimized independently for each analyte and IS. Multiple reactions monitoring the m/z transitions were used for the quantitative analysis of AMD (m/z 646.1/201.1), DEA (m/z 618.1/547.0), and ethopropazine (m/z 313.1/114.1). The LC/MS/MS method achieved lower limits of quantification: 50 ng/ml for AMD and 5 ng/ml for DEA. Analytical reproducibility was judged to be ± 10.2% in the middle of the calibrated range of concentrations. The Pierce BCA protein assay kit (Thermo Scientific, Rockford, IL) was used to determine protein concentration in the liver homogenates. Assessment of cytotoxicity in vitro. The murine hepatoma cell line Hepa1c1c7 purchased from American Type Culture Collection (Manassas, VA) was used to assess cytotoxicity in vitro. Hepa1c1c7 cells were maintained in Dulbecco's modified Eagle's medium (Invitrogen, Carlsbad, CA) with 1% antibiotic-antimycotic (Invitrogen) and 10% heat-inactivated fetal bovine serum (SAFC Biosciences, Lenexa, KS) in 75-cm 2 tissue culture flasks at 37°C in a humidified atmosphere of 95% air and 5% CO 2 . Cells were plated in 96well plates at 15,000 cells per well and allowed to attach for 8 h before medium was replaced. Various concentrations of AMD, DEA, and/or TNF were added to designated wells, and cells were incubated under maintenance conditions. Twenty-four hours later, lactate dehydrogenase (LDH) activity released into the culture medium was measured using the Cytotox-One Homogeneous Membrane Integrity Assay (Promega, Madison, WI). The percent LDH release was calculated as LDH in supernatant/(LDH in supernatant þ LDH in cell lysate). Lysate LDH was determined after addition of Triton to lyse the cells. Statistical analysis. The results are expressed as means ± SEM. One-way or two-way ANOVA was applied as appropriate; Tukey's method was employed as a post hoc test. Grubb's test was used to detect outliers. At least three biological repetitions were performed for each experiment. The p value < 0.05 was set as the criterion for statistical significance. Serum ALT activity did not increase from treatment with either LPS or AMD alone (Fig. 1) . In the AMD/LPS group, administration of AMD at 16 or 20 h before LPS resulted in significant serum ALT activity increase, whereas AMD injected 2-12 h before LPS failed to cause this response. The 16-h interval between AMD and LPS treatments was selected for future studies. Neither AMD alone nor LPS alone affected ALT activity at any of the doses tested. In rats cotreated with AMD and LPS, serum ALT activity was dependent on both AMD ( Fig. 2A) and LPS (Fig. 2B) doses. Significant increases in ALT activity were observed with AMD doses 300 mg/kg (p < 0.05) and with LPS doses 1.2 3 10 6 EU/kg (p < 0.05). 400 mg/kg and 1.6 3 10 6 EU/kg were selected as AMD and LPS doses for subsequent studies, respectively. In the time course study, the serum activities of both ALT and AST were measured as markers for hepatocellular injury (Figs. 3A and B). Neither AMD nor LPS alone affected serum ALT activities at any time examined. For the AST activity, LPS alone had no effect at any time examined, and AMD alone caused a slight increase at 2, 4, and 10 h. Significant elevation of serum ALT and AST activities were only observed in AMD/ LPS cotreatment, and the increases started between 4 and 6 h after LPS administration and continued to increase through 10 h. At 10 h after LPS administration, serum activities of ALP and GGT and concentration of bile acids were measured as indicators of cholestatic injury (Figs. 4A-C). Only AMD/LPS cotreatment caused significant increases in these serum markers, whereas AMD or LPS treatment alone had no effect. All the saline-treated control rats were free of liver lesions (Fig. 5A) . No microscopic evidence of hepatic pathology was found in four of the six LPS-treated rats (Fig. 5B ). Liver sections from two of the LPS-treated rats had a few small foci of midzonal hepatocellular necrosis with an associated neutrophilic influx. In contrast, a widespread, mild-to-marked fibrinopurulent capsulitis was present in the liver sections from all the AMD-treated rats (Fig. 5C ). This was characterized by a thickening of the hepatic capsule due to edema and a conspicuous inflammatory exudate comprising mainly neutrophils, lesser numbers of mononuculear cells, and various amounts of amorphorous proteinaceous material. This fibrinopurulent exudate was also often scattered along the outer peritoneal surface of the capsule. Focal areas of subcapular hepatocellular necrosis were occasionally associated with the capsulitis. The most profound hepatic histopathology was found in animals treated with both AMD and LPS (Fig. 5D ). All these rats had a mild-to-marked fibrinopurulent capsulitis with occasional subcapsular necrosis similar to that found in the AMD-treated rats, but in addition, all these animals had conspicuous areas of midzonal hepatocellular necrosis. The latter lesion ranged from widely scattered focal areas of necrosis in midzonal regions to widespread hepatocellular necrosis with coalescence of affected midzonal and occasionally centriacinar regions (bridging necrosis). Accumulations of neutrophils (inset, Fig. 5D ) were present in all these necrotic regions. Periportal regions were spared of AMD/LPS treatment-related injury. Rats were treated with AMD (0-400 mg/kg, ip) and 16 h later with saline or LPS (1.6 3 10 6 EU/kg, iv). (B) Rats were treated with AMD (400 mg/kg, ip) or its vehicle and 16 h later with LPS (0-1.6 3 10 6 EU/kg, iv). Serum ALT activity was measured at 10 h after LPS administration for both data sets. # indicates significantly different from respective groups not given LPS; * indicates significantly different from respective group not given AMD. p < 0.05, n ¼ 3-14. Accumulation of AMD AMD and DEA concentrations in rat serum and liver homogenates were determined at various times after LPS administration (Figs. 6A-D). From 2 to 10 h after LPS or saline administration (i.e., 18-28 h after AMD administration), the serum and tissue concentrations of AMD and DEA were unaffected by LPS cotreatment. The average serum concentrations of AMD and DEA were 1100 and 128 ng/ml, respectively; and the average liver concentrations of AMD and DEA were 157 and 53 ng/mg protein, respectively. Serum TNF concentration was measured at 2 and 4 h after LPS administration (Fig. 7) . AMD by itself had no effect on serum TNF concentration. At 2 h after LPS, the serum TNF concentrations in rats treated with AMD/LPS or with vehicle/ LPS were similar. However, by 4 h after LPS, the concentration of TNF in serum of AMD/LPS-treated rats was significantly greater than that in vehicle/LPS-treated rats. Hepa1c1c7 cells were exposed to AMD or DEA, and 24 h later, cytotoxicity was assessed by measuring LDH activity released into the culture medium (Figs. 8A and B). Both AMD and DEA caused concentration-dependent LDH release. Significant cytotoxicity was observed with AMD concentrations greater than 20 ug/ml and DEA concentrations greater than 7 ug/ml. Addition of TNF (3 ng/ml) did not cause cytotoxicity alone but significantly potentiated the cytotoxicity of AMD and DEA. Etanercept is a soluble TNF receptor construct that inactivates TNF. Etanercept (8 mg/kg, sc) injected 1 h before LPS inhibited the biological activity of TNF in rats and was not hepatotoxic by itself (Tukov et al., 2007; Zou et al., 2009a) . The same treatment was used in this study. AMD/LPS cotreatment increased serum ALT activity, and etanercept significantly attenuated this increase (Fig. 9A) . Changes in serum ALT activity were supported by histological Rats were treated with AMD (400 mg/kg, ip) or vehicle and 16 h later with LPS (1.6 3 10 6 EU/kg, iv) or saline. They were examined 2, 4, 6, or 10 h after LPS injection. Activities of (A) ALT and (B) AST in serum were measured. # indicates significantly different from respective groups not given LPS; * indicates significantly different from respective group not given AMD. p < 0.05, n ¼ 4-9. AMD HEPATOTOXICITY DURING MODEST INFLAMMATION examination of H&E-stained liver sections: the severity and frequency of necrotic foci were markedly reduced in rats cotreated with etanercept (Fig. 9B) . Since its introduction in Europe in 1962, amiodarone has been associated with idiosyncratic hepatotoxicity (Lewis et al., 1989) . A linear correlation between serum AMD concentration and serum ALT activities has been established (Pollak and You, 2003) ; however, in that study, the ALT values did not exceed 3 3 ULN and were not considered to be clinically significant. For the cases of severe liver injury caused by intravenous amiodarone loading, patients' serum ALT activities were up to 10-206 3 ULN. There was usually a 24-72 h delay between the initial loading of AMD and the onset of elevation in ALT activities; and in many cases, the ALT activity returned to normal after a few days of continuation of maintenance dosing (Rätz Bravo et al., 2005) . Accordingly, it is hard to draw a simple linear relationship between the magnitude or frequency of severe hepatotoxicity and serum AMD concentration. An effort to establish a model for AMD-induced liver injury in healthy rodents was unsuccessful. Neither short-term, large dose nor long-term, small dose administration of AMD led to observable liver damage (Young and Mehendale, 1989) . The cause of severe AMD hepatotoxicity is more likely to be a combination of AMD and other factors, e.g., inflammatory Various concentrations of (A) AMD or (B) DEA were added to cultures of Hepa1c1c7 cells together with TNF (3 ng/ml) or saline. Twenty-four hours after treatment, LDH activity released into the culture medium was measured. The percent LDH release was calculated as described in ''Materials and Methods'' section. * indicates significantly different from respective groups not given TNF; # indicates significantly different from respective groups not given AMD or DEA. p < 0.05, n ¼ 3. episodes. In the present study, regardless of cotreatment with LPS, the serum concentration of AMD was about 1100 ng/ml, which is very close to the steady-state serum concentration of AMD in human patients (1500 ng/ml) under long-term oral amiodarone therapy (Pollak et al., 2000) . Our findings support that at this clinically relevant concentration of AMD in serum, hepatotoxicity can be induced by a concurrent inflammatory episode related to LPS exposure. Previous studies in rodents have suggested a possible association between inflammation and liver injury for several drugs associated with human IADRs, including chlorpromazine (Buchweitz et al., 2002) , ranitidine , diclofenac (Deng et al., 2006) , trovafloxacin (Shaw et al., 2007) , sulindac (Zou et al., 2009b) , and halothane (Dugan et al., 2010) . The results of the present study expand these findings and demonstrate that a nonhepatotoxic dose of AMD is rendered hepatotoxic when acute inflammation is triggered by LPS administration. Acute increases in serum markers for hepatocellular and cholestatic injury were found in rats cotreated with AMD/LPS. These resemble the clinical hepatic chemistry changes in human idiosyncrasy during AMD therapy (Rätz Bravo et al., 2005) . The midzonal and bridging necrosis and infiltration of inflammatory cells in AMD/LPS are also consistent with some of the histological changes found in human patients (Babatin et al., 2008; Rätz Bravo et al., 2005) . However, the histological characteristics in people with AMD-induced liver injury were variable. Different patterns of hepatocellular necrosis, such as midzonal, centrilobular, bridging, and panlobular, have been reported (Lewis et al., 1989) . Genetic differences, concurrent medications, and even different origins of inflammation might account for these varied responses; nevertheless, the AMD/LPS interaction model in rats mimics important aspects of AMD-induced IADRs in human patients. The timing of AMD and LPS dosing in this model was important for the development of severe liver damage. A minimal interval of 16 h was required for AMD and LPS to interact to induce liver injury. When LPS was injected within 16 h after AMD, no liver injury was observed. Absorption, distribution, metabolism, and clearance as well as toxicological actions could contribute to this timing requirement. The elimination of AMD is primarily through hepatic metabolism and biliary excretion, and its half-life in plasma is very long (55 days in humans) (Pollak et al., 2000) . Accordingly, the loss of AMD due to elimination within 16 h is minimal. AMD has a dose-dependent effect on the respiratory chain and b-oxidation in the mitochondria (Fromenty et al., 1990a, b) , and it can also affect the function of lysosomes and other acidic organelles (Stadler et al., 2008) . The 16-h interval may be required for AMD to distribute into the liver, accumulate in organelles, and sensitize hepatocytes to interact with LPS or its downstream cytokines. In other drug/LPS models, there is also a dependence on the temporal relationship between administrations of drug and LPS, and the time interval needed for a maximal hepatotoxic response is different for different drugs (Shaw et al., 2007; Zou et al., 2009b) . This time interval requirement might help to explain the low frequency of IADRs in human patients: i.e., only when the inflammatory episode happens at a certain time during drug therapy would idiosyncratic hepatotoxicity occur. In rats, AMD is deethylated by cytochromes P450 (CYPs) 3A4, 1A1, 2D1, and 2C11 in the liver (Elsherbiny et al., 2008) . DEA, the major metabolite, is three-to fivefold more toxic than AMD to cultured HepG2 cells (Waldhauser et al., 2006) and to primary rat hepatocytes (Gross et al., 1989) . In our treatment of Hepa1c1c7 cells, a similar trend was observed (Fig. 8) . The administration of LPS affects the expression and activities of CYPs in rats (Sewer et al., 1997) . This raised the possibility that LPS might potentiate the toxicity of AMD by increasing its metabolism to DEA. To evaluate this, serum and liver concentrations of AMD and DEA were measured with LC/ MS/MS. The average serum concentration of DEA in AMDtreated rats was 128 ng/ml, which is about 1/10 of the serum AMD concentration. This ratio is commonly seen in the serum after an intravenous loading dose of AMD, both in people (Ha et al., 2005) and rats (Shayeganpour et al., 2008) . Neither the serum nor the liver concentration of AMD or DEA was affected by LPS cotreatment, suggesting that neither AMD accumulation nor DEA generation was affected by LPS. Intratracheal instillation of AMD in vivo or exposure of alveolar macrophages to AMD in vitro led to TNF production (Futamura, 1996; Reinhart and Gairola, 1997) . AMD treatment also increased TNF production by alveolar macrophages on LPS stimulation (Punithavathi et al., 2003) . In the present study, treatment of rats with AMD alone did not cause serum TNF elevation, but it did increase the concentration of TNF in serum of LPS-treated rats. These results suggest that the increased appearance of TNF in AMD/LPS-cotreated rats was probably not an additive effect; rather, AMD appeared to potentiate the production or diminish the clearance of TNF caused by LPS. The concentration of TNF in serum increases rapidly in LPS-treated rats, peaks at around 1.5-2 h, and then returns to basal levels at around 6 h (Tukov et al., 2007) . In the present study, the concentration of TNF around the peak time (i.e., 2 h) in LPS-cotreated rats was not affected by AMD, but the TNF concentration was greater in AMD-cotreated rats at a later time (4 h). These data suggest that AMD prolonged the elevation in TNF caused by LPS administration. This seemingly small difference in TNF concentration was shown to be critical to liver pathogenesis in another drug/LPS model of liver injury involving trovafloxacin (Shaw et al., 2009b) . Accordingly, it is possible that prolongation of the LPS-induced TNF response is a critical event across models of LPS-drug interaction. The importance of TNF in the AMD model was explored by preventing its binding to cellular receptors with etanercept. Etanercept pretreatment reduced hepatotoxicity, indicating that TNF has an important role in AMD/LPS-induced liver injury. Because no liver injury was observed after treatment with LPS AMD HEPATOTOXICITY DURING MODEST INFLAMMATION alone, the large TNF peak caused by LPS was not hepatotoxic by itself; however, this amount of TNF could be critical for the induction of hepatocellular injury by potentiating the toxic effect of AMD and/or DEA. Signaling from an activated TNF receptor can lead to lysosomal leakage, mitochondrial damage, and caspase activation (Wullaert et al., 2007) . All three of these events were also found in AMD and DEA cytotoxicity in vitro (Agoston et al., 2003; Spaniol et al., 2001) . Further support for a critical role for TNF came from our in vitro study in Hepa1c1c7 cells in which TNF increased the cytotoxicity of both AMD and DEA. As a proximal proinflammatory cytokine, TNF can also contribute to liver damage by inducing downstream inflammatory events, such as coagulation activation and neutrophil activation (Shaw et al., 2009b; Tukov et al., 2007) . The etanercept treatment herein reduced the ALT activity to half of the level seen in the absence of this inhibitor, whereas the same dose of etanercept reduced the ALT activity almost to control level in trovafloxacin/LPS and sulindac/LPS models (Shaw et al., 2007; Zou et al., 2009a) . This suggests that other factors induced by LPS might act in parallel with TNF in the AMD/LPS model. In other models of potentiation of xenobiotic toxicity by LPS, factors such as neutrophils (Luyendyk et al., 2005) , the coagulation system (Shaw et al., 2009a) , and prostanoids (Ganey et al., 2001) play important roles, and these factors might be relevant in the model presented here. In summary, AMD was rendered hepatotoxic in rats in the presence of a coexisting inflammatory stress induced by LPS. AMD/LPS-cotreated rats developed liver pathology and blood chemistry changes that resemble AMD-induced idiosyncratic hepatotoxicity in human patients. LPS did not interact with AMD by changing the metabolism or distribution of AMD. AMD enhanced the increase in plasma TNF concentration caused by LPS, and neutralizing TNF reduced liver injury from AMD/LPS coexposure. Moreover, TNF potentiated the cytotoxicity of both AMD and DEA in vitro. These findings add support to the idea that inflammatory stress can interact with IADR-associated drugs to cause liver injury by a mechanism involving TNF and suggest that a similar mode of action might apply to several drugs that cause idiosyncratic hepatotoxicity in humans. National Institutes of Health (R01DK061315).

Polyvalent DNA Vaccines Expressing HA Antigens of H5N1 Influenza Viruses with an Optimized Leader Sequence Elicit Cross-Protective Antibody Responses

Highly pathogenic avian influenza A (HPAI) H5N1 viruses are circulating among poultry populations in parts of Asia, Africa, and the Middle East, and have caused human infections with a high mortality rate. H5 subtype hemagglutinin (HA) has evolved into phylogenetically distinct clades and subclades based on viruses isolated from various avian species. Since 1997, humans have been infected by HPAI H5N1 viruses from several clades. It is, therefore, important to develop strategies to produce protective antibody responses against H5N1 viruses from multiple clades or antigenic groups. In the current study, we optimized the signal peptide design of DNA vaccines expressing HA antigens from H5N1 viruses. Cross reactivity analysis using sera from immunized rabbits showed that antibody responses elicited by a polyvalent formulation, including HA antigens from different clades, was able to elicit broad protective antibody responses against multiple key representative H5N1 viruses across different clades. Data presented in this report support the development of a polyvalent DNA vaccine strategy against the threat of a potential H5N1 influenza pandemic.

The continuous spread of highly pathogenic avian influenza Type A (HPAI) H5N1 viruses in avian species across multiple continents and frequent reports of human H5N1 infection in China and Southeast Asia highlight the threat of a potential flu pandemic in the human population. At the same time, H5N1 viruses have grown into genetically and antigentically diversified viruses. Based on phylogenetic analysis of hemagglutinin (HA) protein gene sequences, at least 10 clades of H5N1 viruses (clades 0-9) have been identified [1, 2, 3, 4, 5] . Recent studies have further assigned these viruses into four major antigenic groups (A-D) [3] . HPAI H5N1 viruses from more than one clade have caused human infection since 1997. A key component in the global strategy to prepare for and control any pending influenza pandemic is the development of an effective vaccine. Several versions of inactivated as well as live attenuated H5N1 vaccines have been tested in humans and showed an overall good safety and immunogenicity profile mainly by using a clade 1 H5N1 virus (A/Vietnam/1203/04) as the vaccine strain per recommendations by the World Health Organization (WHO) [6, 7, 8] . Given that the majority of the world's human population is naïve to H5N1 influenza, two immunizations are needed to achieve desired levels of protective immune responses against H5N1 in contrast to the annual seasonal flu vaccine which requires only one immunization, presumably due to the priming effects by either exposure to circulating H1, H3 or Type B influenza viruses in humans or history of prior seasonal flu vaccination. The likely requirement of two immunizations in conjunction with the genetic complexity of H5N1 viruses, as evidenced by their separation into multiple subgroups, makes it difficult to prepare for the timely production of a sufficient number of doses of H5N1 vaccines in the event of an H5N1 pandemic; therefore, supplemental strategies are needed. As shown by our previously published report [9] and confirmed by other recent studies [10] , a DNA prime-inactivated vaccine boost is highly effective in eliciting higher protective immune responses than using either DNA or inactivated flu vaccine alone. Therefore, it may be possible to use DNA vaccines as the first dose of immunization that can be given either long before the pandemic (pre-pandemic vaccination) or shortly after the outbreak, to reduce the burden on the production of inactivated vaccines at the time of the outbreak. Furthermore, DNA vaccines can be stockpiled for a long period of time, which makes this method even more attractive. One key issue that needs to be analyzed for the above strategy is the cross reactivity between DNA vaccines expressing H5 HA antigens from different clades. It is critical to first optimize the immunogenicity of H5 HA DNA vaccines and then to test how much cross protection can be achieved with optimized H5 HA DNA vaccines. In the current report, we constructed DNA vaccines to express wild type HA antigens without mutations at the HA1 and HA2 cleavage site from four key H5N1 strains that have caused major human infection: HK/156/97 (clade 0), VN/1203/ 04 (clade 1), Ind/5/05 (clade 2.1), and Anhui/1/05 (clade 2.3). Rabbit sera immunized with these HA antigens were examined for their protective antibody responses against either homologous or heterologous H5N1 viruses. Our results demonstrated an imperfect cross-reactivity profile for the protective antibody responses among these four viruses. A polyvalent formulation including three different H5 HA DNA vaccines was able to produce broad protective antibody responses with high titers against these key H5N1 isolates. Information learned from this study should facilitate the selection of candidate H5N1 vaccines to form polyvalent H5N1 DNA vaccines as part of the global strategy to prevent and control a potential avian flu pandemic. One of the key findings from our previous study was that HA antigens from H1 and H3 serotypes had different structure preferences in order to elicit optimal protective antibodies [11] . Two of the HA antigen designs used in that study were also included in the current study to identify the optimal design for the H5 serotype HA antigens: one used the wild type HA antigen insert (H5.wt), which has the exact same amino acid sequences found in the natural viral isolate, and the other used a truncated HA antigen insert (H5.dTM), which removed the transmembrane (TM) and intracellular segments of the HA2 domain ( Fig. 1) . In addition, a third HA antigen insert was created (H5.tPA), in which a human tissue plasminogen activator (tPA) sequence replaced the original wild type leader sequence from the HA antigen (Fig. 1) . The third HA antigen design was adopted because the H5.dTM design also used a tPA leader sequence and the H5.tPA insert served as a control for the H5.dTM insert to understand the role of the tPA leader when it is incorporated as the only change in the design from the original wild type HA antigen insert. In order to maximize the immunogenicity of HA DNA vaccines as shown in previous studies [9, 11] , HA genes used in the current study were also codon optimized and chemically synthesized. In addition, the HA gene sequences used in the current study express intact HA amino acid sequences at the cleavage site between HA1 and HA2 (PQREXRRKKRQG) of HA proteins in highly pathogenic H5N1 viruses [12, 13, 14] . This cleavage was shown to be important for the pathogenesis of H5N1 viruses [12, 15] . For inactivated H5 serotype flu vaccines, these residues were removed to improve the safety profile of such vaccines for both manufacturing and mass immunization purposes [7] [16, 17, 18] . Expression and immunogenicity of different forms of DNA vaccines coding for the HA antigen of a 1997 H5N1 influenza Hong Kong isolate The first set of H5 HA DNA vaccines was produced by cloning codon optimized HA genes based on the amino acid sequences of HA antigen from an H5N1 influenza isolate A/HongKong/156/ 97, which was responsible for the first outbreak of H5N1 avian influenza in humans in 1997, into a DNA vaccine vector [19] . The expression of HA antigens from three different H5-HK DNA vaccines was examined in transiently transfected 293T cells ( Fig. 2A) . Regardless of whether the natural HA leader or tPA leader was used, HA proteins expressed from two full length H5-HK DNA vaccine constructs (HA-HK.wt and HA-HK.tPA) were detected in cell lysate but not in the supernatant, suggesting they are mainly cell-associated. In contrast, truncation of the Cterminal segment, including the removal of the TM domain in the H5-HK.dTM DNA vaccine construct, was able to significantly increase the secretion of HA protein (detected in supernatant) ( Fig. 2A) . H5-HK HA proteins expressed by all of three HA DNA vaccine designs were able to be cleaved into HA1 and HA2 subunits ( Fig. 2A) . New Zealand White (NZW) rabbits were used in the current study to produce large quantities of sera for both binding antibody and functional antibody analyses. Animals were immunized with one of the three H5-HK HA DNA vaccines (individually) via gene gun. Positive antibody responses were elicited in immunized rabbit sera against the HK HA antigen and levels of such responses increased with repeated immunizations while the negative control rabbit group that received empty DNA vector did not have HAspecific antibody responses (Fig. 2B ). As measured by both temporal and peak-level antibody responses, there was no difference in the ability of the three forms of H5-HK HA DNA vaccines to elicit H5 HA-specific antibody responses ( Fig. 2B and 2C) . However, functional antibody analyses with the rabbit anti-HA immune sera showed a very different picture when these sera were further analyzed by either the hemagglutination inhibition (HI) or microneutralization (MN) assays. All three forms of H5-HK HA DNA vaccines induced protective antibody responses against the autologous wild type virus A/HongKong/483/97, but levels of protective antibodies were different among sera induced by different designs of H5-HK HA DNA vaccines. The DNA vaccine with the full length HA insert under the tPA leader sequence (H5-HK.tPA) elicited consistently higher HI and MN antibody titers when compared to the other two forms of HK-HA inserts, the full length HA with a natural leader sequence (H5-HK.wt), and the transmembrane region (TM) truncated HA (H5-HK.dTM). The difference was statistically significant (p,0.05) between H5-HK.tPA and H5-HK.dTM sera based on the HI assay (Fig. 2D ) and between H5-HK.tPA and H5.HK.wt sera based on the MN assay (Fig. 2E) . Expression and immunogenicity of different forms of DNA vaccines coding for the HA antigen of an H5N1 isolate A/VietNam/1203/04 In order to rule out that the above finding was not only unique to this H5N1 virus isolate from Hong Kong in 1997, similar designs of HA inserts were produced by using a codon optimized HA DNA gene from the H5N1 strain A/VietNam/1203/04, a well-studied representative isolate for the H5N1 viruses [20, 21] . The pattern of H5-VN HA expression was similar to that of H5-HK HA antigens. Only cell-associated HA antigens were detected with H5-VN.wt and H5-VN.tPA constructs in contrast to that identified with the H5-VN.dTM, which had HA antigen expression in both cell lysate and supernatant fractions (Fig. 3A) . Rabbits were immunized with the electroporation method as previously reported [22] . Similar to H5-HK DNA plasmids, binding antibody responses, as measured by ELISA, showed similar levels among sera elicited by the three H5-VN DNA vaccines with different HA gene insert designs (Fig. 3B) . However, functional antibodies, as measured by HI and MN antibody analyses, revealed again that the H5-VN.tPA design induced the highest levels of functional antibody responses ( Fig. 3C and 3D ). In the case of functional antibody responses against the wild type virus A/VietNam/1203/04, the differences between H5-VN.tPA and the other two forms were statistically significant by both HI and MN assays (p,0.05 or p,0.01). In order to ensure that the difference in protective antibody responses between sera elicited by H5-VN.wt and H5-VN.tPA was not the result of repeated immunizations, sera collected after one or three immunizations were also measured (Fig. 4) . Pseudotyped viruses expressing VN HA antigen were used to measure the neutralizing antibody activities in rabbit sera with less immunizations. The strength of protective antibodies was measured at two levels: inhibition concentrations that can block either 50% (IC50) or 90% (IC90) of virus infection to target cells. Both measurements showed that H5-VN.tPA-elicited rabbit sera had significantly higher titers of neutralizing antibody activities than the wild type H5 HA design, especially when using the more stringent IC90 as a cut-off (p,0.05 or p,0.01) (Fig. 4) . Sensitivity to deglycosylation treatment for HA antigens expressed by different forms of H5-VN HA DNA vaccines Additional studies were conducted to ask if glycosylation of HA has been affected with the use of a different leader sequence which may influence the immune responses. Our previous study with a hepatitis B surface antigen suggested that post-translational modifications including glycosylation may affect the immunogenicity of DNA vaccine delivered antigens [23] . The Asn (N)-linked glycosylation of influenza HA proteins are essential for virus infectivity and vaccine immunogenicity [24, 25] . Based on sequence analysis, there are 8 to 9 N-linked glycosylation sites (Asn-X-Ser/ Thr) in H5N1 HA proteins. We next investigated whether the HA antigens expressed by different designs of H5 HA inserts in the above DNA vaccinations have similar levels of N-linked glycosylation. The HA antigens expressed in 293T cell lysate transfected with the H5-VN.wt, H5-VN.tPA or H5-VN.dTM DNA vaccines and HA antigen expressed in the 293T cell supernatant transfected with the H5-VN.dTM DNA vaccine were analyzed for their susceptibility to PNGase treatment, which can cleave any type of Nlinked sugar including complex oligosaccharide structures resulting from the maturation of high mannose moieties during transport of the glycoprotein through the Golgi. Treatment with PNGaseF allowed for the removal of N-linked glycosylations, as shown by the reduction of apparent molecular weight of HA0, HA1, and HA2 species in Western blot analysis for HA proteins expressed in cells transfected by these H5-VN HA DNA vaccines ( Figure S1 ). The molecular weight reduction patterns for HA0, HA1, and HA2 antigens were the same between cells transfected by H5-VN.wt and H5-VN.tPA DNA vaccines, suggesting that HA antigens expressed by these two HA DNA vaccines were similarly glycosylated. A smaller molecular weight HA2 antigen was observed in cells transfected with the H5-VN.dTM DNA vaccine, presumably due to the truncated size of HA2 domain in this particular HA insert design; however, PNGaseF treatment also led to a proportional reduction of molecular weight for truncated HA2 protein in both supernatant and cell lysate preparations. The only unique finding is that the cell-associated HA antigens in H5-VN.dTM transfected cells showed a high level of heterogeneity and a small portion of the HA1 proteins was not fully deglycosylated by PNGaseF treatment, reflecting the continued presence of different forms of glycosylated HA proteins in H5-VN.dTM transfected cells. Otherwise, the overall glycosylation pattern, as probed by deglycosylation treatment, was very similar among HA antigens produced by three different types of H5 HA DNA vaccines. Cross-protective antibody responses induced by individual DNA vaccines expressing HA antigens from key H5N1 viral isolates Based on the above results, additional HA DNA vaccines with the HA.tPA insert design were produced by using codon optimized HA genes that encode the HA proteins from H5N1 viral strains A/Anhui/1/2005 and A/Indonesia/5/2005, both have caused human infection in recent years [26, 27] . Given the circumstance that H5N1 influenza antigen drifts have occurred since the first human outbreak in Hong Kong in 1997, it would be important to determine if H5 HA vaccines developed based on H5N1 viruses isolated at different epidemic time points can induce cross antibody responses against other H5N1 viruses. One set of experiments was conducted to understand the cross protection between paired H5N1 HA antigens. Rabbits were immunized with individual H5-HK.tPA, H5-VN.tPA, and H5-AH.tPA DNA vaccines and rabbit immune sera were examined for HA antigen-specific antibody responses. H5 HA-specific antibody responses against H5 HA antigens from different viruses were analyzed by ELISA and the potential cross-protective antibody responses against different H5N1 viruses were evaluated by HI and MN assays. H5-HK (A/HK/156/97), a clade 0 H5N1 isolate, and H5-VN (A/VN/1203/04), a clade 1 H5N1 isolate, represent H5N1 viruses isolated from the first human outbreak in Hong Kong in 1997 and a subsequent outbreak in Vietnam in 2004, respectively. Results shown in Fig. 5 indicate that H5-HK.tPA DNA vaccine-immunized rabbit sera showed high antibody responses recognizing both the autologous H5-HK HA antigen and the heterologous H5-VN HA antigen. However, the HA-specific IgG titers against the autologous H5-HK antigen were higher than those observed against the heterologous H5-VN antigen (p,0.05). Conversely, the H5-VN.tPA DNA vaccine elicited high level HA-specific IgG responses against autologous H5-VN and heterologous H5-HK HA antigens although the overall titers against its autologous H5-VN HA antigen may be higher (not statistically significant). Protective HI and MN antibody responses induced by H5-HK.tPA and H5-VN.tPA DNA vaccines were further compared against either A/HK/483/97 or A/VN/1203/04 wild type viruses (Fig. 5 ). HI and MN titers were present in both HK.tPA and H5-VN.tPA DNA vaccine-immunized rabbit sera against Similar analysis was conducted with rabbit immune sera elicited by H5-VN and H5-AH HA DNA vaccines (Fig. 6 ). H5-AH (A/ Anhui/1/05), a clade 2.3 H5N1 isolate, represents the H5N1 virus isolated from a human outbreak in China in 2005 [26] . The cross reactivity between H5-VN and H5-AH immune sera and viruses was very similar to that observed above between H5-HK and H5-VN immune sera and viruses. For binding antibody responses, H5-VN rabbit immune sera had significantly higher recognition to its autologous H5-VN HA antigen than the heterologous H5-AH HA antigen (p,0.05). H5-AH rabbit immune sera elicited higher antibody responses recognizing the autologous H5-AH HA antigen than the heterologous H5-VN HA antigen (not statistically significant) (Fig. 6) . For functional antibodies, both HI and MN analyses showed preference for H5-VN and H5-AH rabbit immune sera against their respective autologous wild type viruses, A/ VN/1203/04 and A/Anhui/1/05 (p,0.05 or p,0.01) (Fig. 6 ). Cross-protection by a polyvalent DNA vaccine formulation expressing HA antigens from three representative H5N1 viral isolates The above results indicate that while antibody responses against one H5N1 HA antigen or virus may well cross-react with another H5N1 HA antigen or virus, the levels of antibody responses against the autologous antigen or virus were always higher. Given the uncertainty regarding which H5N1 virus may ultimately cause a pandemic H5N1 outbreak, it is important to develop a vaccine strategy that can maximize the protection efficacy against a wide spectrum of H5N1 viruses from different clades (or subtypes). It is possible that a consensus HA antigen, or a structurally-optimized HA antigen design, can cover different H5N1 viruses at various levels of protective efficacy but there is no actual data showing that such HA antigen can achieve the maximum protective antibody responses against several H5N1 viral isolates from different clades. One alternative approach is to produce a polyvalent HA formulation by including multiple H5N1 HA DNA vaccines in one injection to produce an immune sera that can induce the highest antibody response against a wide range of H5N1 viral isolates. The pseudotyped virus system, developed in recent years and widely used in leading influenza studies, provides high sensitivity in detecting functional antibody responses against influenza HA antigens, and at the same time, eliminates the influence of other influenza viral gene products since a common viral backbone is used for different HA pseudotyped viruses [28, 29] . It is an ideal system as a high-throughput assay for multiple serum samples against a wide range of viruses. As shown in Fig. 6 , rabbit sera elicited by either the 3-valent HA DNA vaccine formulation (H5-VN+H5-AH+H5-IN) or the monovalent H5 HA DNA vaccines (HK, VN, AH, and IN) were tested for their antibody responses against three pseudotyped viruses expressing H5-VN, H5-AH, or H5-IN HA antigens. The matched monovalent rabbit sera consistently showed the highest functional antibody responses against autologous pseudotyped viruses (Fig. 7A-7C) . However, the 3-valent serum was the only one that showed high level antibody responses against all three pseudotyped viruses while one non-matched monovalent rabbit serum could neutralize one or two viruses but not all three, further confirming the hypothesis that a polyvalent HA formulation is capable of protect against multiple H5N1 viruses. According to phylogenetic analysis, H5N1 viruses can be divided into 10 clades (0-9). Since 1997, humans have mainly been infected by H5N1 viruses from clades 0, 1, and 2, although there have also been reports of infection by clade 7 virus [5] . Clade 2 is the most complicated in its genetic evolution and has been further divided into five subclades (2.1 to 2.5). In the current study, While progress has been made in reducing the number of required immunizations during vaccination with inactivated H5N1 vaccines by incorporating various adjuvants into the vaccine formulations, a major next-step for H5N1 vaccine research is to determine to what degree the immunity elicited by one H5 avian influenza vaccine (currently, many candidate H5N1 vaccines were developed based on a clade 1 virus (A/VietNam/1203/2004)) can cross-protect against H5N1 viruses from other clades. Unlike human Type A influenza viruses (H1 or H3 serotypes), any potential pandemic caused by an H5N1 virus will be of avian origin and, in theory, any of the current known H5N1 avian viruses may jump to the human population leading to the next pandemic. Therefore, a systemic examination on the cross-protection among HA antigens from different clades is needed for strategic planning to determine whether more than one H5N1 vaccine is needed based on the analysis of protection profiles, and if so, what particular viral strains should be selected to provide the maximum breadth of protection. DNA vaccination is an attractive strategy to provide relatively quick and straightforward production of vaccines against an influenza pandemic when the demand for such vaccines suddenly increases. However, a key issue surrounding the use of DNA vaccines is their low immunogenicity in humans. In recent years, the success of the prime-boost strategy has greatly enhanced the utility of DNA vaccination for future human applications [10, 30] . At the same time, optimization of the design of antigen inserts based on the uniqueness of each antigen (a process of ''antigen engineering'') [31, 32] can also play a key role. Results included in the current report indicated that the tPA leader sequence and the C-terminal transmembrane domain/ cytoplasmic region of H5 HA both contribute to better functional antibody responses in H5.tPA DNA vaccines when compared to H5.wt and H5.dTM DNA vaccines. The above findings were different from our previous results on protective antibody responses induced by differently designed flu H1 and H3 HA DNA vaccines [11] . In this previous study, only the full length H1.wt but not transmembrane truncated H1.dTM induced high level HI and MN responses against H1 virus while both the full length H3.wt and truncated H3.dTM induced similar HI and MN responses against H3 virus [11] . These results provide a strong indication that the HA antigens from different influenza A subtypes (H1, H3 and H5) may have different preferences for antigen structure designs in order to generate optimal protective antibody responses. Studies were conducted in this report to identify the mechanism responsible for better protective antibodies in rabbit immune sera elicited by the H5.tPA HA insert design but the exact mechanism is currently unclear. First, we tested whether a higher level of HA antigen expression was produced with the tPA-leader design. As shown in Fig. 2A and 3A, antigen expression levels between WTleader design and tPA-leader design were similar, thus excluding this possibility. Next, we asked whether there is increased secretion of the HA antigen due to the use of the tPA leader. However, as shown in Fig. 2A and 3A , there is no major detectable level of secreted HA antigens in supernatant for either the WT-design or tPA-design. Furthermore, the dTM design did have a higher level of secretion due to the deletion of transmembrane and intracellular portion of HA protein but did not elicit better protective antibody responses. Finally, a study was conducted to examine the possible role of post-translational processing, such as a change in glycosylation, of HA the antigen, which may affect the antigen processing pathway, as we previously reported with a hepatitis B surface antigen DNA vaccine [33, 34, 35] . However, Figure S1 showed that there is no major difference between WTdesign and tPA-design after de-glycosylation treatment. Therefore, it is very likely that HA antigen expressed with the tPA leader may be more effective in eliciting conformational antibodies. This hypothesis is supported by two pieces of evidence. First, there was no difference in the levels of binding antibodies as measured by ELISA, which indicated that there was no difference in the general immunogenicity between WT and tPA leader designs; only a difference in the functional antibody was observed. Second, the HA.dTM DNA vaccine design also used the tPA leader but did not have better functional antibodies, proving that the proper folding of the HA antigen in the presence of a tPA leader is important and is dependent on the presence of an intact HA2 domain. It is possible that such an HA antigen conformation is part of a trimer structure of HA since the HA2 domain is involved in the formation of HA trimers. Only antibodies against the trimer structure are more functionally relevant to block the trimer form of HA spikes on viral particles. By using the optimal H5.tPA HA insert design, studies in this report further demonstrated that there are good levels of cross protection by one H5 HA DNA vaccine against multiple H5N1 viruses from different clades. It is well documented that cross protection among H5N1 viruses can be detected [33, 34, 35] . Clade 1 H5N1 vaccines (VN) cross protect against both clade 1 and clade 2 (Indonesia) viruses in ferrets with the use of a strong adjuvant [36] . Using live attenuated cold adapted (ca) viruses expressing HA and NA from 1997, cross protection was observed against the late H5 virus from 1997 to 2005 [37] and ferrets [37] and ferrets [38] and mice [39, 40] and mice [39, 40] . However, as shown in the current study, not all H5 HA vaccines can elicit the same levels of cross protective antibodies, and more significantly, maximum levels of protective antibodies were usually detected against the autologous viral isolates. Given the knowledge that protective anti-flu antibody responses in humans are much There were 3 rabbits/group in Vector, HK, VN, AH, and Ind groups, and 4 rabbits/group in 3-valent groups. Rabbit sera tested by the pseudotyped NAb assays were collected at 2 weeks after the 4 th DNA immunization. '','' denotes below detection level. The arrow denotes neutralization against the autologous H5 pseudotyped virus. The statistical differences between the testing group and the autologous neutralization group or 3-valent group are indicated by ''*'' when the p value was less than 0.05. doi:10.1371/journal.pone.0028757.g007 lower than in experimental animals, cross protection may not be very high in humans with one randomly selected H5 HA vaccine. In the current set of studies, it was encouraging to observe that the polyvalent H5 HA DNA vaccine was able to elicit high level protective antibody responses against multiple key H5N1 viruses. Such a polyvalent flu DNA vaccine can be used for stockpiling against a potential H5N1 pandemic even before information is available on which viral isolates may cause a human outbreak. At the same time, there are alternative approaches including the use of consensus HA antigen designs to achieve a broad coverage of various viral strains (personal communication with David Weiner). It will be interesting to compare the relative efficacy between polyvalent and consensus HA DNA vaccines in their abilities to elicit protective antibody responses. One unique advantage of the polyvalent formulation is its flexibility; one alternative HA antigen can replace or be added to the earlier polyvalent formulation in the event that a new strain of virus becomes a threat while it will be necessary to re-design the whole consensus HA insert in order to allow for broader coverage. While HI titers against heterologous virus (cross-clade) were significantly reduced compared to HI titers against homologous virus in the current study, the heterologous virus titers were generally above 1:100, and in humans, a HI titer of 1:40 has been associated with protection [41] , and so all of the constructs might be protective after multiple immunizations. At the same time, the titers observed in mice may not be the same as in humans. While HI titers have been associated with protection, H5 HA DNA vaccines described in the current report were not tested for protection against challenge and so there is the possibility that only some of these constructs may not protect against clinical disease or lethal infection. No matter the design of DNA vaccines that may be used, recent studies have indicated that DNA priming immunization is effective as part of the prime -boost strategy for flu vaccine applications. In addition to DNA prime-inactivated flu vaccine boost [9, 10] , a study published in 2011 further demonstrated that DNA primelive attenuated flu vaccine boost was equal to or more effective than twice immunization with the live attenuated flu vaccine against the H5N1 viruses based on antibody responses and viral clearance in immunized ferrets [42] . Since live attenuated vaccines are considered the most immunogenic form of vaccines, it is impressive to observe that one time DNA prime was able to achieve the same priming effect as a live attenuated flu vaccine. In this particular study, the H5-VN.tPA DNA insert was used as part of the collaboration with the manufacturer of live attenuated H5N1 flu vaccine. Based on the results published in the current report and other recent similar studies, H5N1 HA DNA vaccines evaluated in the current study should be included in the design of human studies to understand whether results reported here can be reproduced in humans when they are used as part of DNA prime, either individually or as part of the polyvalent HA DNA formulation. The finding from such studies will be very useful in the identification of simple yet powerful approaches to develop vaccines against major influenza pandemics. preference of Homo sapiens. The less optimal codons in HA genes were changed to the preferred codons in mammalian systems to promote higher expression of the HA proteins, as previously described [11] . These codon optimized HA genes were chemically synthesized by Geneart (Regensburg, Germany) with added restriction enzyme sites of PstI and BamHI for subcloning purpose immediately upstream of the start codon and downstream of the stop codon, respectively. For either H5-HK or H5-VN HA DNA vaccines, three versions of codon optimized HA gene inserts were cloned into DNA vaccine vector pSW3891 [43] . For the first version, the full length H5-HK or H5-VN HA gene insert (568 aa,) with their natural HA leader sequences subcloned, individually, into the pSW3891 vector at the PstI and BamHI sites, designated as H5-HK.wt or H5-VN.wt DNA vaccine constructs. For the second version, the HA natural leader sequence (the first 15 aa at N-terminus for both H5-HK and H5-VN) was replaced by a human tissue plasminogen activator (tPA) leader sequence. The H5-HK and H5-VN HA gene inserts coding for aa 16-568 were PCR amplified from the full length codon optimized H5 HA genes using the following primers: H5-HA-opt-1 (gtcgctccgctagc GACCAGATCTGCATC-GGCTAC) and H5-HA-opt-2 (agtcacggatcc TCAGATGCA-GATCCGGCACTG). The individual H5-HK or H5-VN HA gene was cloned into the pSW3891 vector at the NheI and BamHI sites downstream of the tPA leader sequence and designated as H5-HK.tPA or H5-VN.tPA DNA vaccine constructs. For the third version, the HA natural leader sequence was replaced by a tPA leader sequence and the transmembrane (TM) and cytoplasmic region (37 aa at the C-terminus) of H5 and HA was deleted for both H5-HK and H5-VN. The truncated H5-HK or H5-VN HA genes were PCR amplified from the full length codon optimized H5-HK or H5-VN HA gene using primer pairs: H5-HA-opt-1 and H5-HA-opt-4 (agtcac ggatccTCACTGGTAG-GTGCCCATGCTCTC), or H5-HA-opt-1 and H5-HA-opt-8 (agtcacggatccTCACTGGTAGATGCCGATGCTTTC), respectively. The truncated H5-HK or H5-VN gene insert was individually cloned into the pSW3891 vector at the NheI and BamHI sites downstream of the tPA leader sequence and designated as H5-HK.dTM or H5-VN.dTM. For the H5-AH HA DNA vaccine, the construct with the full length HA under tPA-leader sequence was made as described above. Each individual DNA vaccine plasmid was prepared from Escherichia coli (HB101 strain) with a Mega purification kit (Qiagen, Valencia, CA) for both in vitro transfection and in vivo animal immunization studies. NZW rabbits (,2 kg body weight) were purchased from Millbrook Breeding Labs (Amherst, MA) for immunogenicity studies, and housed in the Department of Animal Medicine at the University of Massachusetts Medical School in accordance with IACUC approved protocol. The rabbits (3 rabbits/group) were immunized with a Helios gene gun (Bio-Rad) at the shaved abdominal skin as previously reported [44] with a total of 36 mg H5 HA DNA vaccine plasmid or vector control plasmid at each immunization. DNA immunizations were given at weeks 0, 2, 4, 8. Serum samples were taken prior to the first immunization and 2 weeks after each immunization for study of H5 HA-specific antibody responses. This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the University of Massachusetts Medical School's Institutional Animal Use and Care Committee (IACUC) (Protocol: A-1674). All surgery was performed under sodium pentobarbital anaesthesia, and all efforts were made to minimize suffering. Transient expression of the HA antigens from various HA DNA vaccine constructs were verified by Western blot analysis. HA DNA vaccine constructs were first transfected into the human embryonic kidney 293T cells using the calcium phosphate precipitation method. Briefly, 2610 6 293T cells at 50% confluence in a 60 mm dish were transfected with 10 mg of plasmid DNA, and a total of 3 ml supernatant and 100 ml of cell lysate were harvested 72 hours later. Equal amounts of each transiently expressed HA antigen (10 ng of protein in 10 ml) were loaded for the SDS-polyacrylamide gel electrophoresis (SDS-PAGE) under denatured conditions, then transferred onto PVDF membranes (Bio-Rad, Hercules, CA). After being blocked overnight at 4uC in blocking buffer (0.2% I-block, 0.1% Tween-20 in 16PBS), the membranes were incubated with a 1:500 dilution of rabbit sera immunized with HA DNA vaccines for 30 min followed by washes. Then, the membranes were incubated with alkaline phosphatase-conjugated goat anti-rabbit IgG at 1:5000 dilution for 30 min. Following washes, the signals were detected using a chemiluminescence-based Western-Light Kit (Tropix, Bedford, MA). To analyze the N-linked glycosylation of H5 HA antigens expressed by various forms of H5 HA DNA vaccines, the HA antigens expressed from 293T cells [45, 46] were treated with PNGaseF (New England BioLab, Beverly, MA) [47, 48, 49] . Briefly, the HA proteins were first denatured at 100uC for 10 min in glycoprotein denaturing buffer and then chilled on ice. Following addition of G7 reaction buffer, the deglycosylation enzyme cocktail was added and incubated reaction at 37uC for 4 hours. Either mock-treated or deglycosylated HA samples were subjected to SDS-PAGE and Western blot analysis was performed as described above. ELISA was conducted to measure HA-specific antibody (IgG) responses in immunized rabbits and mice. The 96-well flat-bottom plates were coated with 100 ml of ConA (50 mg/ml) for 1 hour at room temperature, and washed 5 times with PBS containing 0.1% Triton X-100. Subsequently, the plates were incubated overnight at 4uC with 100 ml of transiently expressed HA antigen at 1 mg/ ml. After being washed 5 times as above, the plates were then blocked with 200 ml/well of blocking buffer (5% non-fat dry milk, 4% Whey, 0.5% Tween-20 in PBS at pH7.2) for 1 hour. After five washes, 100 ml of serially diluted rabbit or mouse serum was added in duplicate wells and incubated for 1 hour. After another set of washes, the plates were incubated for 1 hour at 37uC with 100 ml of biotinylated anti-rabbit or anti-mouse IgG (Vector Laboratories, Burlingame, CA) diluted at 1:1000 in Whey dilution buffer (4% Whey, 0.5% Tween-20 in PBS). Then, 100 ml of horseradish peroxidase-conjugated streptavidin (Vector Laboratories) diluted at 1:2000 in Whey buffer was added to each well and incubated for 1 hour. After the final washing, the plates were developed with 3,39,5,59 Tetramethybenzidine (TMB) solution at 100 ml per well (Sigma, St. Louis, MO) for 3.5 minutes. The reactions were stopped by adding 25 ml of 2 M H 2 SO 4 , and the plates were read at OD 450 nm. The end titration titer was determined as the highest serum dilution that has an OD reading above twice of that from the negative control serum. Influenza A viruses of the H5N1 A/HongKong/483/97 (H5N1), A/Viet Nam/1203/04 (H5N1), A/Anhui/1/2005 (H5N1), and A/ Indonesia/5/2005 (H5N1) were grown in the allantoic cavity of 10day-old embryonated hen eggs at 37uC for 26 to 40 h. Allantoic fluid pooled from multiple eggs was clarified by centrifugation and frozen in aliquots at 270uC. The 50% egg infectious dose (EID50) for each virus stock was calculated by the method of Reed and Muench following serial titration in eggs. All experiments with HPAI viruses were conducted under Biosafety Level 3 containment. Hemagglutination-inhibition (HI) assay HI assays were performed by standard methods [50] . Briefly, the assay was performed using, 0.5% v/v fowl or horse [30] red blood cells, 4 HA unit of reference H5N1 virus (A/HongKong/ 483/97, A/VietNam/1203/04, A/Anhui/1/2005, A/Indonesia/ 5/2005) and specific sera treated with receptor destroying enzyme. The HI titer was defined as the highest dilution of the serum able to inhibit hemagglutination. MN assays was performed as described previously [51] . In brief, influenza virus containing 100 TCID 50 was incubated with equal volume of two-fold dilutions of the specific heat-inactivated serum overnight at 37uC in a 5% CO2 humidified atmosphere for 1 hr. After the incubation, 100 ml virus-serum samples were added to a 96-well plate containing Madin Darby Canine Kidney (MDCK) cell monolayer and incubated for 5 days at 37uC and 5% CO2. The microneutralization titer was defined as the highest dilution of serum that neutralized 100 TCID 50 of virus in MDCK cell [52, 53] cultures (as detected by the absence of cytopathic effects). The MN assays were conducted in two different labs: 1) National Microbiology Laboratory, Public Health Agency of Canada, and 2) Beijing Institute of Microbiology and Epidemiology, according to the viruses available. The recombinant lentiviral vectors expressing a luciferase reporter gene were produced as previously described [28, 54, 55] . To produce H5N1 pseudotyped viruses, 293T cells [45, 46] (5610 6 cells plated the day before) were transfected with 13.43 mg of pNL 4-3.Luc.R-E-(NIH AIDS reference and reagent program), 1.2 mg of H5-HA-wt DNA vaccine plasmid and 0.3 mg of N1-NA plasmid using 75 mg of polyetheleneimine transfection reagent. Supernatants were harvested 48 hours later, frozen at 280uC and then standardized by infectivity in 293A cells using a luciferase-based TCID50 measurement. For neutralization assays, serum samples (5 ml) were heat inactivated at 56uC for 30 minutes, then threefold serially diluted in culture medium in flat-bottomed microtiter plates. Pseudotyped H5N1 virions were then added to the plates at 200 TCID50/well and incubated for 1 hour at 37uC. After incubation, 293A cells were trypsinized and added to each plate at a dilution of 1610 4 cells per well. Following a 48-hour incubation, plates were developed using a luciferase assay system (Promega). Values averaged from triplicate wells were then used to determine IC50 based on wells that displayed 50% reduction in infection as compared to control wells containing virus plus pre-immune sera.

ELM—the database of eukaryotic linear motifs

Linear motifs are short, evolutionarily plastic components of regulatory proteins and provide low-affinity interaction interfaces. These compact modules play central roles in mediating every aspect of the regulatory functionality of the cell. They are particularly prominent in mediating cell signaling, controlling protein turnover and directing protein localization. Given their importance, our understanding of motifs is surprisingly limited, largely as a result of the difficulty of discovery, both experimentally and computationally. The Eukaryotic Linear Motif (ELM) resource at http://elm.eu.org provides the biological community with a comprehensive database of known experimentally validated motifs, and an exploratory tool to discover putative linear motifs in user-submitted protein sequences. The current update of the ELM database comprises 1800 annotated motif instances representing 170 distinct functional classes, including approximately 500 novel instances and 24 novel classes. Several older motif class entries have been also revisited, improving annotation and adding novel instances. Furthermore, addition of full-text search capabilities, an enhanced interface and simplified batch download has improved the overall accessibility of the ELM data. The motif discovery portion of the ELM resource has added conservation, and structural attributes have been incorporated to aid users to discriminate biologically relevant motifs from stochastically occurring non-functional instances.

Short linear motifs (SLiMs, LMs or MiniMotifs) are regulatory protein modules characterized by their compact interaction interfaces (the affinity and specificity determining residues are usually encoded between 3 and 11 contiguous amino acids (1)) and their enrichment in natively unstructured, or disordered, regions of proteins (2) . As a result of limited intermolecular contacts with their interaction partners, SLiMs bind with relatively *To whom correspondence should be addressed. Tel: +49 (0) 6221 3878398; Fax: +49 (0) 6221 387517; Email: gibson@embl-heidelberg.de low affinity (in the low-micromolar range), an advantageous attribute for use as transient, conditional and tunable interactions necessary for many regulatory processes. Due to the limited number of mutations necessary for the genesis of a novel motif, SLiMs are amenable to convergent evolution, functioning as a driver of network evolution by adding novel interaction interfaces, and thereby new functionality, to proteins. This evolutionary plasticity facilitates the rapid proliferation within a proteome, and as a result, motif use is ubiquitous in higher eukaryotes. SLiMs play an important role for many regulatory processes such as signal transduction, protein trafficking and post-translational modification (3, 4) . Their importance to the correct functionality of the cell is also reflected by the outcome of motif deregulation. For example, point mutations in SLiMs have been shown to lead severe pathologies such as 'Noonan-like syndrome' (5) , 'Liddle's syndrome' (6) or 'Retinitis pigmentosa' (7) . Furthermore, mimicry of linear motifs by viruses to hijack their hosts' existing cellular machinery plays an important role in many viral life cycles (8) . However, despite their obvious importance to eukaryotic cell regulation, our understanding of SLiM biology is relatively limited, and it has been suggested that, to date, we have only discovered a small portion of the human motifs (9) . Several resources are devoted to the annotation and/or detection of SLiMs [Prosite (10), MiniMotifMiner (11) and Scansite (12) ]. Here, we report on the 2012 status of the Eukaryotic Linear Motif database. The ELM initiative (http://elm.eu.org) has focused on gathering, storing and providing information about short linear motifs since 2003. It was established as the first manually annotated collection of SLiM classes and as a tool for discovering linear motif instances in proteins (13) . As it was mainly focused on the eukaryotic sequences, it was termed the Eukaryotic Linear Motif resource, usually shortened to ELM. The ELM resource consists of two applications: the ELM database of curated motif classes and instances, and the motif detection pipeline to detect putative SLiM instances in query sequences. In the ELM database, SLiMs are annotated as 'ELM classes', divided into four 'types': cleavage sites (CLV), ligand binding sites (LIG), sites of posttranslational modification (MOD) and subcellular targeting sites (TRG) ( Table 1) . Currently, the ELM database contains 170 linear motif classes with more than 1800 motif instances linked to more than 1500 literature references (Table 1 ). Each class is described by a regular expression capturing the key specificity and affinity determining amino acid residues. A regular expression is a computer-readable term for sequence annotation and is used by the ELM motif detection pipeline to scan proteins for putative instances of annotated ELM classes. The search form for sequence input is shown in Figure 1 , while the results page showing the putative and annotated instances is illustrated in Figure 2 . The ELM resource is powered by a PostgreSQL relational database for data storage and a PYTHON web framework for data retrieval/visualization. The main tables within the database contain information about ELM classes, ELM instances, sequences, references, taxonomy and links to other databases [the database structure is described in greater detail in (14) ]. Since the last release (14) , 24 new ELM classes have been added to the ELM database (Table 1 ) and several more have been updated. One of the newly annotated motif classes is the AGC kinase docking motif (LIG_AGCK_PIF), consisting of three distinct classes. It is present in the non-catalytic C-terminal tail of AGC kinases that constitute a family of serine/threonine kinases consisting of 60 members that regulate critical processes, including cell growth and survival. Deregulation of these enzymes is a causative factor in different diseases such as cancer and diabetes. The motif interacts with the PDK1 Interacting Fragment (PIF) pocket in the kinase domain of AGC kinases. It mediates intramolecular binding to the PIF pocket, serving as a cis-activating module together with other regulatory sequences in the C-tail. Interestingly, in some kinases the motif also acts as a PDK1 docking site that trans-activates PDK1, which itself lacks the regulatory C-tail, by interacting with the PDK1 PIF pocket. PDK1 in turn will phosphorylate and activate the docked kinase. Other novel classes (Table 2) include phosphodegrons, which are important mediators of phosphorylation-dependent protein destruction, and the LYPxL motif, which is involved in endosomal sorting of membrane proteins but is also implicated in retrovirus budding. Annotated ELM instances serve as representative examples of the respective ELM class. They are also invaluable for the computational analysis and classification of motifs (15) . Therefore, special emphasis has been put on the curation of more than 500 novel ELM instances (in 40 different classes) by scanning and annotating more than 400 articles. The number of protein databank (PDB) entries annotated have been increased to 195 (Table 1 ), meaning that for 10% of all instances there is a 3D Figure 1 . ELM start page. The user can submit a query sequence to the motif detection pipeline either as UniProt accession number or in FASTA format. Filtering criteria such as taxonomic range or cellular compartment should be activated to limit the resulting list of SLiM instances. protein structure annotated, giving more detailed information about the biological context of the respective motif. The ELM website at http://elm.eu.org can be used in two ways: first, as a front-end to explore the ELM database of curated ELM classes and instances, and second, to run the motif detection pipeline to detect putative SLiM instances in query sequences. Both interfaces have been improved with the most notable changes listed below. The database user interface, having been stable for many years, has been overhauled and replaced by a novel interface introducing several new features ( Figure 1 ). Up-to-date web technologies have been used to improve the general user experience: the PYTHON framework DJANGO (http://www.djangoproject.com) dynamically creates and serves all HTML pages, while JavaScript was used to make the whole site more interactive and thus improve the user experience. In particular, the ELM detail pages (Figure 3) , which hold the most (18) . The lower part contains the annotated and putative ELM instances for the given protein sequence (Epsin1, UniProt accession Q9Y6I3). The background is colored according to the structural information available. Each box represents one ELM instance, the color of which indicates the likelihood that this instance is functional: grey instances are buried within structured regions, while shades of blue represent instances outside of structured regions and hint on sequence conservation, with pale blue representing weak sequence conservation and dark blue indicating strong sequence conservation. Red ellipses or boxes mark instances that are annotated in the query sequence or a homologous sequence, respectively. important information about each ELM class including references, regular expression, taxonomic distribution and gene ontology terms (Table 3) , have been updated by annotating the protein domain interacting with the respective motif. Where available, a 3D model of representative protein databank structures of linear motif interactions was added to the ELM detail page ( Figure 3 , top right). To cope with the increasing amount of annotated classes as well as instances, a novel query interface was introduced to assist the user in finding information of interest. The ELM browser (Figure 4 ) now features a search interface for free text search. In addition, the search results can also be filtered and reordered using buttons (Figure 4 , left side) and table headers, respectively, and be downloaded as tab-separated values (TSV). Further, improvements to the ELM database include revising the experimental methods used for annotation by using a standardized methods vocabulary [in sync with PSI-MI ontology (16, 17) ]. A candidate page has been introduced to display novel ELM classes that have not yet been annotated in detail or are currently undergoing annotation. We invite researchers to send us their feedback and expert opinion on these classes and to contribute novel motif classes that will be added to the candidate page and ultimately be turned into full ELM classes ( Figure 5 ). Minimum requirements are at least one literature reference as well as a short description. In addition, a draft regular expression or a 3D structure showing the relevant interaction would also be helpful. Currently, the number of possible ELM classes on this candidate list (awaiting further annotation) exceeds the number of completely annotated classes, indicating the great demand for further annotation. The ELM motif detection pipeline scans protein sequences for matches to the regular expressions of annotated ELM classes ( Figure 2 ). The query output combines these putative instances with information from the database (annotated ELM instances) as well as predictions from different algorithms/filters. The ELM resource employs a structural filter (18) to highlight and mask secondary structure elements, as well as SMART (19) to detect protein domains. Furthermore, an additional disorder prediction algorithm (IUPred) (20) has been included to predict ordered/disordered regions within the protein. IUPred uses a cutoff of 0.5 to classify a sequence region as either structured or disordered, with values above this threshold corresponding to disorder, highlighted in green background and lower values indicating structured regions, displayed in red background in the output graph. Disorder and domain information is combined by Motifs, present in proteins in several repeats, which mediate binding to the hydrophobic cleft created by subdomains 1 and 3 of G-actin LIG_Actin_WH2_2 LIG_Actin_RPEL_3 The AGCK docking motif mediates intramolecular interactions to the PDK1 Interacting Fragment (PIF) pocket, serving as a cis-activating module LIG_AGCK_PIF_2 LIG_AGCK_PIF_3 IAP-binding motifs are found in pro-apoptotic proteins and function in the abrogation of caspase inhibition by inhibitor of apoptosis proteins in apoptotic cells LIG_BIR_III_1 LIG_BIR_III_2 LIG_BIR_III_3 LIG_BIR_III_4 Motif binding to the dorsal surface of eIF4E LIG_eIF4E_2 A proline-rich motif binding to EVH1/WH1 domains of WASP and N-WASP proteins LIG_HCF-1_HBM_1 The background coloring to highlight structured regions within the protein, which allows inspection of SLiMs that reside at domain boundaries and emphasizes motifs in disordered regions. The conservation of linear motifs can help in assessing the functional relevance of putative instances, with functional instances showing higher overall sequence conservation than non-functional ones (21) . Therefore, sequence conservation of the query protein is calculated using a tree-based conservation scoring method (22) and highlighted in the graphical output. Here, lighter shades of blue represent low conservation while dark blue shading corresponds to high-sequence conservation. The actual conservation score can be inspected by moving the mouse over the respective ELM instance (Figure 2) . The functionality of linear motifs can be modulated by modifications such as phosphorylation (23, 24) . To enable the user to investigate phosphorylation data in the context of putative linear motif instances, phosphorylation annotations from the Phospho.ELM resource (25) have been added to the graphical output (Figure 2, top row) . The phosphorylated residues are highlighted in different colors (serine: green, threonine: blue, tyrosine: red); each phosphorylation site is linked to a page showing detailed information about the respective modification site from the manually curated data set of the Phospho.ELM resource. The importance of the short linear motifs in virus-host interactions makes the ELM resource an important tool for the viral research community. For example, Cruz et al. (26) analyzed a protein phosphatase 1 (PP1) docking motif in 'protein 7' of transmissible gastroenteritis virus using the ELM class LIG_PP1. This conserved sequence motif mediates binding to the PP1 catalytic subunit, a key regulator of the cellular antiviral defense mechanisms, and is also found in other viral proteomes, suggesting that it might be a recurring strategy to counteract the hosts' defense against RNA viruses by dephosphorylating eukaryotic translation initiation factor 2a and ultimately ribonuclease L. To reflect our increasing awareness of viral motifs (8), special focus has been attributed to the annotation of viral instances in the ELM database: in the latest release, more than 200 novel ELM instances found in 84 different viral taxons have been added. The notion of viruses abusing existing SLiMs in their hosts is demonstrated by viral instances being annotated alongside instances in their hosts' proteins. For example, the ELM class LIG_PDZ_Class_1 contains 12 instances in human proteins but has recently been expanded with 5 instances from 5 different human pathogenic virus proteins. . ELM instances browse page. A full-text search (here, search term used was 'AP2', filtering for 'true positive' instances in taxon 'Homo sapiens', yielding 58 instances) assists in finding annotated instances. A search can be restricted to a particular taxonomy or instance logic (top) or ELM class type (buttons on the left). The list can also be exported to TSV or FASTA format for further processing. The importance of SLiMs is further corroborated by the occurrence of pathologies that are caused by mutations that either mutate existing linear motifs or create novel linear motifs (of undesired function) (27) . Examples include 'Usher's syndrome' (28) , 'Liddle's Syndrome' (6) or 'Golabi-Ito-Hall Syndrome' (29) . The developmental disorder 'Noonan Syndrome' can be caused by mutations in Raf-1 that abrogate the interaction with 14-3-3 proteins mediated by corresponding SLiMs and thereby deregulate the Raf-1 kinase activity (30) (the Raf-1 protein sequence features two LIG_14-3-3_1 binding sites that are annotated at 256-261 and 618-623 in the ELM resource). A related disease, 'Noonan-like Syndrome', is caused by an S to G mutation at position 2 of the SHOC2 protein, creating a novel myristoylation site (annotated as ELM class MOD_NMyristoyl). This irreversible modification results in aberrant targeting of SHOC2 to the plasma membrane and impaired translocation to the nucleus upon growth factor stimulation (5) . More information about the implication of short linear motifs on diseases is collected at http://elm.eu.org/infos/diseases.html. By providing a high-quality, manually curated data set of linear motif classes with experimentally validated SLiM instances, the ELM database has proven to be invaluable to the community: small-scale (single protein) analyzes benefit from the detailed annotation of each ELM class in attributing novel features to proteins of interest. By using in vitro and in vivo studies, von Nandelstadh et al. (31) could validate a PDZ class III motif, detected by ELM at the carboxy terminus of myotilin and the FATZ (calsarcin/myozenin) families. This evolutionarily conserved carboxy-terminal motif mediates binding to PDZ domains of ZASP/Cypher and other Enigma family members (ALP, CLP-36 and RIL) and disruption of these interactions results in myofibrillar myopathies (32) . Additionally, ELM annotations can contribute to high-throughput screenings (33) as well as development of novel algorithms (34) (35) (36) , methods (37) and databases (38) . Furthermore, the highly curated data of the ELM resource are used as a benchmarking data set to evaluate the accuracy of prediction algorithms (21, 39, 40) . For any such analysis, the user should be aware that many matches to ELM regular expressions are false positives. Before conducting experiments based on ELM results, it is strongly advisable to check if a motif match is conserved, exposed in a cell compartment in which the motif is known to be functional. The ELM resource applies several filters to provide the user with such information that should ideally also be supported by the experimental evidence. The importance of SLiMs is highlighted by the growing number of instances with relevance to diseases or viruses. Yet, despite their importance and abundance, our understanding of linear motifs is still limited. This is mainly owing to the fact that they are still quite difficult to predict computationally and to investigate experimentally (3, 41, 42) . By better understanding the biology of linear motifs, we hope to increase our insight into diseases and viruses (and vice versa). The ELM resource tries to aid the researcher in the search for putative SLiM instances by providing a feature-rich toolset for sequence analysis. Consequently, with the aforementioned additions and changes, we hope that the ELM resource continues to be a valuable asset to the community.

Epithelial Cells Derived from Swine Bone Marrow Express Stem Cell Markers and Support Influenza Virus Replication In Vitro

The bone marrow contains heterogeneous population of cells that are involved in the regeneration and repair of diseased organs, including the lungs. In this study, we isolated and characterized progenitor epithelial cells from the bone marrow of 4- to 5-week old germ-free pigs. Microscopically, the cultured cells showed epithelial-like morphology. Phenotypically, these cells expressed the stem cell markers octamer-binding transcription factor (Oct4) and stage-specific embryonic antigen-1 (SSEA-1), the alveolar stem cell marker Clara cell secretory protein (Ccsp), and the epithelial cell markers pan-cytokeratin (Pan-K), cytokeratin-18 (K-18), and occludin. When cultured in epithelial cell growth medium, the progenitor epithelial cells expressed type I and type II pneumocyte markers. Next, we examined the susceptibility of these cells to influenza virus. Progenitor epithelial cells expressed sialic acid receptors utilized by avian and mammalian influenza viruses and were targets for influenza virus replication. Additionally, differentiated type II but not type I pneumocytes supported the replication of influenza virus. Our data indicate that we have identified a unique population of progenitor epithelial cells in the bone marrow that might have airway reconstitution potential and may be a useful model for cell-based therapies for infectious and non-infectious lung diseases.

Bone marrow contains a variety of stem cells that include hematopoietic stem cells, mesenchymal stem cells or stromal cells (MSC), and multipotent adult progenitor cells [1] . Many reports using a variety of animal models have demonstrated that bone marrow cells (BMCs) may have a role in the repair and regeneration of injured lung, infarcted myocardium, and damaged bone, tendon and cartilage [2, 3, 4, 5, 6, 7, 8] . BMCs cultured in vitro can differentiate into type I, II, and basal and airway epithelial cells and express the cystic fibrosis transmembrane conductance regulator (CFTR) protein [9] . BMCs have been shown to improve survival and attenuate lung inflammation in bleomycin-and endotoxin-induced lung injury [8, 10, 11, 12] . Following infusion of BMCs in animal models, these cells have been identified as type I and II alveolar epithelial cells, endothelial cells, fibroblasts, and bronchial epithelial cells [12] . However, precise identity of specific subpopulation of BMCs that engraft in the lung parenchyma and have regenerative potential is still not clear. Recently, Wong and colleagues [13] reported the isolation of progenitor epithelial cells from mouse and human bone marrow. These cells expressed Clara cell secretory protein (Ccsp), a marker of airway progenitor cells [14] , CD45 and mesenchymal markers CD73, CD90, CD105. These cells differentiated into multiple epithelial cell lineages, including type I and II pneumocytes in vitro. Furthermore, these progenitor epithelial cells preferentially homed to naphthalene-injured lung following intratracheal or systemic inoculation. Influenza viruses belong to the family Orthomyxoviridae and cause highly contagious respiratory infections in humans and animals. These viruses cause seasonal epidemics and infrequent pandemics in humans. Seasonal influenza epidemics are responsible for between 200,000 and 500,000 influenza-related deaths each year [15] . Avian influenza viruses caused three human pandemics during the last century. The 2009 pandemic, the first pandemic of 21 st century, was caused by a triple reassortant H1N1 influenza virus of swine lineage [16] . In addition to seasonal and pandemic viruses, highly pathogenic avian influenza (HPAI) H5N1 virus has crossed species barrier to infect humans. As of August 9, 2011, more than 500 human cases with over 300 deaths have been reported worldwide [17] . H5N1 viruses replicate to higher titers in lungs and extra-pulmonary tissues leading to acute respiratory distress syndrome, multiple-organ dysfunction, lymphopenia, and hemophagocytosis [18, 19, 20] . Influenza viruses, therefore, pose a constant public health threat, and it is important to understand its pathogenesis to devise effective control measures. Swine are gaining popularity as a useful large animal model for stem cell therapy for important human diseases or conditions such as myocardial infarction, diabetes, atherosclerosis, traumatic brain injury, retinal damage, and tooth regeneration [21, 22, 23, 24, 25] . Like humans, pigs are an outbred species. As well, they are similar to humans in anatomy, physiology, and immune responses [26, 27, 28, 29] . Additionally, swine can serve as an excellent animal model for influenza virus pathogenesis studies. The clinical manifestations and pathogenesis of influenza in pigs closely resemble to what is observed in humans. Furthermore, the cytokine responses in branchoalveolar lavage (BAL) fluid from swine influenza virus-infected pigs are also identical to that observed in nasal lavage fluids of experimentally infected humans [30] . These observations support that pigs serve as an excellent animal model to study the pathogenesis of influenza virus [31] . In this study, we report the isolation of previously undocumented progenitor epithelial cells in pig bone marrow that expressed Clara cell secretory protein (Ccsp), a marker for lung progenitor cells, and the stem cell markers octamer-binding transcription factor (Oct4) and stage-specific embryonic antigen-1 (SSEA-1). These progenitor cells showed increased self-renewal capacity and expressed epithelial cell markers such as pancytokeratin (Pan-K), cytokeratin 18 (K-18), and occludin. Importantly, these cells expressed receptors for both mammalian and avian influenza viruses and were permissive to infection with these viruses. The progenitor cells differentiated into type I and II pneumocytes and type II pneumocytes also supported replication of influenza virus. These data provide new insights into the pathogenesis of influenza virus. Further, porcine progenitor epithelial cells described here may serve as a useful model in cellular therapy strategies for epithelial diseases. During the culture of BMCs for the isolation of mesenchymal stromal cells (MSC), we observed colonies of epithelial cells surrounded by mesenchymal cells. The colony cells exhibited cuboidal morphology typical of epithelial cells (Fig. 1A) . To characterize epithelial colony cells, we expanded individual colonies in vitro and performed immunocytochemistry by using a panel of stem cell and epithelial cell-specific antibodies. We found that the epithelial colony cells expressed Ccsp, Oct4, and SSEA-1 (Fig. 1B) . We detected the expression of Oct4 in the nuclei, whereas SSEA-1 was mainly found on the cell surface and in the cytoplasm of the progenitor epithelial colony cells (Fig. 1B) . The colony cells expressed epithelial specific markers Pan-K, K-18, and occludin ( Fig. 1C) but not the mesenchymal markers CD29, CD44, and CD90 (data not shown). Differentiation potential of progenitor epithelial cells To address the self-renewal and differentiation potential of these progenitor epithelial cells, individual colonies were isolated from primary cultures and seeded in tissue culture plates precoated with collagen I in 50% epithelial growth medium and 50% MSC conditioned medium (MSC-CM); referred to as epithelial differentiation medium in the text. In epithelial differentiation medium, the cells continued to grow, became flattened, and appeared as thinly spread cell clusters ( Fig. 2A) . By day 5, the expression of pro surfactant protein C (SPC) was detected in the cytoplasm of these flattened cells (Fig. 2B ). These features are consistent with the type II pneumocytes. Also, expression of aquaporin 5 (Aqua5) protein, a marker for type I pneumocytes was detected on expanded cells which were significantly larger than that of the parental primary epithelial colony cells (Fig. 2B ). The expression of SPC or Aqua5 was not detected on primary undifferentiated cells. These results suggest that bone marrow progenitor epithelial cells have the potential to differentiate into type I and II like pneumocytes. Detection of a-2,3and a-2,6-linked sialic acid receptors on progenitor epithelial cells Influenza virus infects cells through binding to cell surface sialic acid receptors. To examine the susceptibility of progenitor epithelial cells to influenza virus, we first evaluated by flowcytometry these cells for the presence of sialic acid receptors. The a-2,3-and a-2,6-linked sialic acid receptors were detected on the surface of progenitor epithelial cells by lectin staining. A majority of progenitor cells (.95%) expressed both a-2,3and a-2,6-linked sialic acid receptors, indicating that viruses of both avian and mammalian lineages might replicate in these cells (Fig. 3A ). After confirming the expression of sialic acid receptors on progenitor epithelial cells, we next examined the susceptibility of progenitor epithelial cells to mammalian and avian influenza viruses. Cells were inoculated with SwIV, AvIV or HuIV at a MOI of 1 for 1 hour (h). As shown in Fig. 3 , infectious viruses were observed in culture supernatants after 24 h after infection and viral titers slightly increased at 36 h. Among the viruses tested, SwIV produced highest cellular lysis (Fig. 3B ) and replicated to highest titers followed by AvIV and HuIV (Fig. 3C) . The production of cytopathic effects required live, infectious virus because cell cytotoxicity was not observed in heat-inactivated virus-infected cultures (data not shown). In addressing whether in vitro differentiated type I and II pneumocytes are susceptible to influenza virus infection, progenitor cells were differentiated to type I and II pneumocytes for 5 days in epithelial differential medium. The differentiated cultures were then exposed to SwIV. Co-immunostaining revealed SwIV replication in differentiated type II pneumocytes as indicated by the presence of viral proteins whereas no viral proteins were detected in differentiated type I pneumocytes (Fig. 4 ). In this study, we report the isolation and identification of previously undocumented stem/progenitor epithelial cells from bone marrow of pigs. These cells expressed the stem/progenitor cell markers Oct4, SSEA-1, and Ccsp, and the epithelial markers Pan-K, K-18, and occludin. Upon culture in epithelial differentiation media, these cells differentiated into type I and II pneumocytes. Most importantly, progenitor cells were targets for mammalian and avian influenza virus replication. We detected the expression of Oct4 in progenitor epithelial cells. Oct4, a member of the POU family of transcription factors, is expressed in pluripotent stem cells such as embryonic stem cells, induced pluripotent stem cells, and lung stem cells where it regulates self-renewal and pluripotency [32, 33] . These Oct4 + colony cells also expressed other stem cell markers; SSEA-1, Ccsp and epithelial markers; pan-K, K-18, and occludin. Clara cells, airways progenitor cells in the lungs which have been involved in lung regeneration and repair, also express Ccsp and cytokeratins [14, 32, 34] . Importantly, these Oct4 + Ccsp + colony cells differentiated into cells possessing Aqua5 and SPC which are markers for type I and II like pneumocytes respectively. Expression of stem cell markers in progenitor epithelial cells may, therefore, be associated with differentiation potential of these cells. The porcine progenitor epithelial cells reported in this study share similarities with recently reported human and mouse cells that both cell types express Ccsp, but unlike human and mouse cells, porcine progenitor epithelial cells did not express mesenchymal markers such as CD29, CD44 and CD90 [13] . Porcine progenitor epithelial cells also expressed stem cell markers; Oct4 and SSEA-1; however, mouse and human cells were not tested for expression of these markers [13] . Similar to human and mouse bone marrow cells, porcine bone marrow progenitor cells expressed both Ccsp and SPC. Ccsp and SPC expressing broncho-alveolar stem cells (BACS) were recently identified in adult mouse lung [35] . Mouse BACS were found to express hematopoietic markers, whereas our porcine bone marrow progenitor epithelial cells lacked the expression of these markers. Importantly, we were able to passage progenitor epithelial cells up to passage 8 suggesting that these cells have extensive self-renewal and proliferation properties. The role of bone marrow stem cells in tissue regeneration including lungs is well established [36, 37, 38, 39] . However, the precise identity of specific subpopulation of BMCs that has tissue regeneration capacity is not known. As reported for human progenitor epithelial cells, porcine Oct4 + Ccsp + colony cells were detected in the plastic-adherent fraction surrounded by mesenchymal cells [40] . Therefore, it is likely that the progenitor epithelial cells were derived from mesenchymal precursors. Previously, bone marrow and even cord blood MSC were shown to have epithelial differentiation potential. Cord blood MSC when cultured in vitro expressed lung-specific markers such as Ccsp, SPC, and CFTR [41] . Intravenous administration of cord blood-MSC into NOD-SCID mice resulted in a low level of airway engraftment of the cord blood-MSC that expressed cytokeratin and CFTR. Similarly, bone marrow MSC differentiated to type I and II pneumocytes in vitro and following intravenous or intratracheal administration, engrafted in recipient lung paren-chyma as type I and II pneumocytes [3, 42] . These observations suggest that bone marrow MSC could serve as precursors of differentiated epithelial cells. Moreover, in this study we observed that progenitor epithelial cells when cultured in 50% MSC-CM were able to differentiate into type I and II pneumocytes further providing evidence that mesenchymal stroma is important for the pluripotency of epithelial colony cells. Future work will be directed to identify the molecular signals provided by the mesenchymal precursors that regulate the differentiation potential of epithelial progenitor cells. Bone marrow cell therapy may be effective in conditions that currently lack effective treatment. Stem cells isolated from different anatomic niches of lungs [34, 35, 43, 44, 45] have been shown to differentiate into multiple epithelial cell types following lung injury. Our lab and Wong and colleagues [13] demonstrated the existence of progenitor epithelial cells in the bone marrow of pigs and humans respectively which are capable of differentiating into type I and II pneumocytes suggesting that these cells will be valuable for lung therapies. Functionally, type I pneumocytes are important for gas exchange, water permeability and the regulation of alveolar fluid homeostasis whereas type II pneumocytes produce lung surfactants that reduce the alveolar surface tension [46] . The isolation and expansion of lung-derived autologous cells from humans is difficult, whereas bone marrow progenitor cells can be easily isolated and expanded. In other preliminary studies, we have isolated progenitor epithelial cells from the lung of pigs. Future in vivo experiments in pigs will be designed to compare the lung repair potential of bone marrow-and lung-derived progenitor epithelial cells following infection and/or LPS-induced lung injury. Excessive virus replication, multi-organ failure, and hyperimmune activation have been detected in humans and laboratory animals infected with HPAI H5N1 and with recreated 1918 pandemic influenza virus [47, 48, 49, 50, 51] . The mechanisms responsible for deterioration of lung function and the loss of capacity for lung repair after influenza virus infection are not well understood. Although influenza virus primarily replicates in lung, virus has been detected in extra pulmonary tissues including bone marrow [52, 53] . Our in vitro findings show that progenitor epithelial cells in bone marrow are permissive to influenza virus suggesting that in the event of influenza virus infection, bone marrow progenitor cells may not be available to home to the injured lung to participate in tissue repair. Additional experiments are needed to confirm the infection of progenitor epithelial cells in vivo and to delineate further their roles in local and lung repair following infection with influenza virus. In conclusion, we have isolated Oct4 + Ccsp + SSEA-1 + expressing progenitor epithelial cells in the bone marrow of swine which are capable of differentiating into type I and II pneumocytes. Additionally, we have demonstrated that these cells serve as targets for influenza virus infection. Further characterization of these progenitor cells may help in understanding the pathogenesis of infectious and non-infectious epithelial diseases and the mechanisms of lung injury repair. Femur bones were obtained from 4-to 6-week old germ-free pigs. BMCs were isolated by previously described methods [54, 55, 56] . Briefly, the tip of each bone was removed and the marrow was harvested by inserting a syringe needle into one end of the bone and flushing with Dulbecco's Modified Eagle's Medium (DMEM; Gibco). The bone marrow cells were filtered through a 70-mm nylon mesh filter (BD, Falcon, USA) and mononuclear cells were obtained by density gradient centrifugation over Ficoll-Hypaque (gradient density 1.077). Cells (1-5610 5 /cm 2 ) were plated in 25 cm 2 cell culture flasks in DMEM containing 10% fetal bovine serum, 2 mm L-glutamine, 1% antibiotic solution (Gibco, USA). Cultures were incubated at 37uC in a humidified atmosphere containing 95% air and 5% CO 2 . The non-adherent cells were removed after 72 h of culture. Individual epithelial colonies surrounded by mesenchymal cells were plucked and were expanded in vitro by culturing in DMEM media. Stem cell and epithelial cell markers on progenitor epithelial cells were detected by immunofluorescence assay (IFA). Epithelial colony cells were fixed in methanol: acetone (1:1) for 3 min at room temperature and then blocked with 3% bovine serum albumin (BSA) for 30 min. Cells were incubated at 4uC with following primary antibodies: rabbit anti-human Oct4 (Santa Cruz Biotechnology), mouse anti-human SSEA-1(Millipore), goat antimouse cc10 (Santa Cruz Biotechnology), mouse anti-human pan cytokeratin, mouse anti-human cytokeratin 18 (Sigma), rabbit anti-human occludin (Invitrogen), mouse anti-human CD90, rat anti-mouse CD44 and mouse anti-pig CD29 (BD Pharmingen). After overnight incubation at 4uC, cells were washed and incubated for 1 h at room temperature with the following respective FITC-labeled secondary antibodies: donkey anti-goat IgG, goat anti-rabbit IgG, and goat anti-mouse IgG (Sigma). Cell nuclei were then counterstained with 49,6-diamidino-2-phenylindole (DAPI). Porcine bone marrow MSC were grown and expanded as described above. When MSC reached at about 80% confluence, medium was aspirated and cells were washed three times with phosphate-buffered saline. Cells were cultured in serum-free medium for 24 h. CM was removed and centrifuged to remove cellular debris and used in epithelial differentiation assays. To analyze differentiation potential of porcine progenitor epithelial cells, cells were seeded in culture dishes pre-coated with collagen I. The cells were cultured in epithelial cell differentiation medium containing 50% epithelial growth media supplemented with bovine pituitary extract (70 mg/ml), human epidermal growth factor (5 ng/ml), insulin (5 mg/ml), and hydrocortisone (0.5 mg/ ml) (MEGM, Lonza, Walkersville, MD) and 50% MSC-CM medium. After 5 days of incubation at 37uC, cells were observed for morphology and cells grown on coverslips were examined for the expression of Aqua5 (marker for type I pneumocytes) and SPC (marker for type II pneumocytes) by using goat anti-human Aqua5 (Santa Cruz Biotechnology) and rabbit anti-human SPC (Millipore) primary antibodies by IFA as described above. Sialic acid receptors on progenitor epithelial cells were detected by flowcytometry [57] . Madin-Darby canine kidney (MDCK) cells have been shown to express both a-2,3-linked sialic acid receptors and a-2,6-linked sialic acid receptors; avian and mammalian influenza virus receptors, respectively. Therefore, MDCK cells were included as positive controls. The cells were incubated with FITC-labeled Maackia amurensis lectin II (MAA) or Sambucus niagra agglutinin (SNA) (EY Laboratories) at 4uC for 30 min in dark and acquired by Accuri C6 flow cytometer (BD Accuri) and analyzed using CFlow plus software (Accuri). Influenza virus strains; swine influenza virus (SwIV; Sw/OH/ 07, H1N1), avian influenza virus (AvIV; Ch/NY/95, H7N2) and human influenza virus (HuIV; Hu/OH/06, H1N1) were propagated in 10-day-old embryonated chicken eggs (AvIV was kindly provided by Dr. S.M. Goyal, Department of Veterinary Population Medicine, University of Minnesota, Saint Paul, MN). Progenitor epithelial cells were infected with SwIV, AvIV and HuIV at a multiplicity of infection (MOI) of 1. After adsorption for 1 h, the cells were washed and fresh medium was added to the cells. At intervals after infection, released viruses in culture supernatants were titrated in MDCK cells [58] . To examine whether differentiated progenitor epithelial cells support influenza virus infection, epithelial colony cells were cultured in collagen I coated dishes for 5 days in epithelial cell differentiation medium. On day 5, cells were infected with SwIV and influenza viral nucleoprotein (NP) was detected at 24 h after infection by IFA using mouse anti influenza A NP (Millipore) as primary antibody and rhodamine-labelled goat anti-mouse (Molecular Probes) as secondary antibody.

Two Birds with One Stone? Possible Dual-Targeting H1N1 Inhibitors from Traditional Chinese Medicine

The H1N1 influenza pandemic of 2009 has claimed over 18,000 lives. During this pandemic, development of drug resistance further complicated efforts to control and treat the widespread illness. This research utilizes traditional Chinese medicine Database@Taiwan (TCM Database@Taiwan) to screen for compounds that simultaneously target H1 and N1 to overcome current difficulties with virus mutations. The top three candidates were de novo derivatives of xylopine and rosmaricine. Bioactivity of the de novo derivatives against N1 were validated by multiple machine learning prediction models. Ability of the de novo compounds to maintain CoMFA/CoMSIA contour and form key interactions implied bioactivity within H1 as well. Addition of a pyridinium fragment was critical to form stable interactions in H1 and N1 as supported by molecular dynamics (MD) simulation. Results from MD, hydrophobic interactions, and torsion angles are consistent and support the findings of docking. Multiple anchors and lack of binding to residues prone to mutation suggest that the TCM de novo derivatives may be resistant to drug resistance and are advantageous over conventional H1N1 treatments such as oseltamivir. These results suggest that the TCM de novo derivatives may be suitable candidates of dual-targeting drugs for influenza.

The first global pandemic of the 21st century was announced by the World Health Organization (WHO) in 2009 due to the worldwide spread of influenza A subtype H1N1 (H1N1/09) [1] . More than 214 countries have reported laboratory confirmed cases, and more than 18,449 deaths have been recorded [2] . Currently, the neuraminidase inhibitor TamifluH (oseltamivir) remains the primary drug prescribed to patients infected with H1N1/09 [3] . However, the emergence of drug resistant viral strains [4] and limited drug administration window [5] exemplifies the need for additional therapies. Important constituents of influenza surface membrane proteins include hemagglutinin, neuraminidase, and the matrix protein 2 (M2) proton channel [6, 7] . Hemagglutinin mediates the binding of viral particles to host cell surface sialic acid and the invasion of viruses into host cell [8] [9] [10] . Neuraminidase is responsible for the cleavage of sialic acid residues to promote the release of progeny viruses [11, 12] . M2 proton channels are critical for viral mRNA incorporation into the virion and virus budding [13] . Over one hundred serological subtypes [14] have been identified through different combinations of the 16 hemagglutinin (H1-H16) and nine neuraminidase groups (N1-N9) currently known. The 3Dstructure of M2 proton channels have recently been solved in both influenza A and B [15, 16] , allowing more in depth studies regarding its biological function and action mechanism [17] [18] [19] . These proteins have been used as targets for rational attempts to design drugs for influenza [20] [21] [22] [23] [24] [25] [26] [27] . The H1N1/09 virus strain is a triple reassortant that contains gene segments from avian, swine and human influenza viruses [28] . In addition to antigenic shift that can lead to fundamental changes in influenza surface antigens, antigenic drift could reduce binding affinity of host antibodies to antigens [29, 30] . A major challenge in influenza vaccine development is the rapid evolution of influenza viruses, causing vaccines to be easily outdated and reformulation necessary each year [31] [32] [33] . Although the H1N1/ 09 virus is susceptible to neuraminidase inhibitors, cases regarding oseltamivir-resistant viruses with neuraminidase mutation (such as H275Y) have been reported [34, 35] . Given that influenza viruses have RNA genomes that are prone to changes, it is imperative to devise new therapies. Much effort has been made to investigate the mechanism and devise alternative drugs against the drugresistance issue of H1N1 [36] [37] [38] [39] [40] . Developing inhibitors that target both H1 and N1 antigens can reduce resistance issues resulting from the mutation of a single target antigen. Computational approaches have been widely applied to molecular biology and medicine [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] . Structure-based methods, including docking and MD simulation, are invaluable tools in The influenza A subtype H1N1 (H1N1/09) pandemic raised public concerns due to drug resistance strains. Drug resistance occurs from conformational changes causing the original drug to lose binding ability and exhibit biological effects. The world's largest TCM Database@Taiwan was employed to screen for potential leads that simultaneously bind to H1 and N1. Three de novo compounds derived from Rosemarinus officinalis and Guatteria amplifolia were identified as having dual binding properties to H1 and N1. Structural analysis indicated that the candidates bind to multiple residues in both H1 and N1. In addition, the de novo derivatives were predicted as bioactive using four different computational models. The compounds are validated as potent dual targeting influenza drug candidates through multiple validations. Key advantages of the candidates include (1) binding to H1 and N1 through multiple amino acids, and (2) not binding to known mutation residues in H1 or N1. Such advantages can reduce drug resistance caused by single point mutations. On a broader context, features important for successful H1N1 drug development are discussed in hopes of providing starting templates for drug development and improvements. drug discovery and design. Computational docking is important for investigating ligand-protein interactions and elucidating binding mechanisms [51] [52] [53] [54] [55] [56] [57] . Since publication of the pioneer paper in 1977 [58] , it has been established that low-frequency motions existing in proteins and DNA can help reveal dynamic mechanisms underlying fundamental biological functions [59] [60] [61] [62] [63] . NMR observation later confirmed such inferences and the findings were applied to medical treatments [64] [65] [66] [67] . In recent years, application of molecular dynamics to investigate internal motions and biological functions of biomacromolecules has opened new frontiers. Vast amounts of information on molecular recognition and binding [68] [69] [70] [71] , conformations or conformational changes [72] [73] [74] [75] , molecular mechanisms of bioactivity and stability [76] [77] [78] [79] , and drug discovery [80] [81] [82] [83] [84] have been found. To understand interaction of drugs with proteins or DNA, consideration should be given not only to the static structures but dynamical information obtained by simulation through a dynamic process. In this regard, both docking and MD simulation were utilized in this study to provide comprehensive analysis protein-ligand interactions under static and dynamic conditions. Much effort has been placed on developing new, effective influenza treatments, but most have focused on neuraminidase or M2 as the target protein [37, 38, [85] [86] [87] . To date, no hemagglutinin inhibitor is available. Traditional Chinese medicine (TCM) has been used extensively for finding effective drugs [88] , and we have successfully designed novel medicinal compounds and identified potential drug leads through traditional Chinese Medicine Database@Taiwan (TCM@Taiwan) [89] . Preliminary studies conducted in this lab show potential for TCM compounds to serve as neuramindase and hemagglutinin inhibitors individually [90] [91] [92] [93] [94] [95] . In view of the current needs for drugs effective against native and mutant H1N1/09 and our promising preliminary results, the present study integrates the concept of ''dual targeting'' with the aforementioned computational tools and TCM in the attempt to identify dual-targeting inhibitors of H1N1 that may be useful for drug development. The experimental procedures and screening results after each filtering step are summarized in Figure 1 . Among the 829 native TCM compounds, 81 docked into both H1 and N1 and were used for de novo evolution (Table S1 ). De novo compounds with dual binding capacities to H1 and N1 were ranked by combined DockScore and the top ten derivatives are listed in Table 1 . Nine of the ten top ranking de novo compounds were derived from Rosmaricine, a natural compound isolated from Rosemarinus officinalis [96] . The remaining de novo compound was based on Xylopine, which is naturally found in Guatteria amplifolia [97] . The top three derivatives, Xylopine_2, Rosmaricine_14 and Rosmaricine_15, have in common a pyridinium addition to their native structure ( Figure 2) . The pyridinium addition could be the main explanation for higher DockScores of these three derivatives compared to their native compounds and the other derivatives. Rosmaricine_14 and Rosmaricine_15 differed by the number of fused rings, but the slight difference in DockScore suggests that addition of an acyclic ring has little influence on binding affinity. Docking of the de novo compounds back to the receptor provides insights to modifications that can be made to modulate or enhance molecular properties and also highlights important protein-ligand interactions. When docked into the N1 protein binding site, Xylopine_2 interacts with Asp151 via a protonated amino group and has pi and hydrogen bond (H-bond) interactions with Trp179 and Glu228, respectively ( Figure 3A ). Rosmaricine_14 ( Figure 3B ) and Rosmaricine_15 ( Figure 3C ), have interactions with Asp151 and Arg293 via the carbonyl group and Glu228 through the 2-aminopyridinium group. TamifluH forms H-bond interactions with Arg156, Arg293 and Arg368, but not with Asp151 or Glu228 ( Figure 3D ). Both Asp151 and Glu228 have been reported as one of the major residues in the N1 ligand binding site [98, 99] . The ability of the de novo derivatives to form interactions with both Asp151 and Glu228 may account for the higher DockScores. Binding of the top three de novo derivatives to H1 site is detailed elsewhere [92] . The ability to bind with important H1 residues Asp103 and Arg238 [100] indicates the dual targeting possibility of the candidates. Figure 4A . All values were within the 95% prediction bands and the r 2 value = 0.8043. The SVM model was constructed using identical molecular descriptors and ligands as the MLR model. The r 2 value of the SVM model was 0.8605 and the correlation between observed and predicted activities of 27 ligands are illustrated in Figure 4B . Table 2 summarizes the pIC 50 values of TamifluH and the top three candidates as predicted by the generated MLR and SVM models. The predicted activity of TamifluH using the generated MLR model (pIC 50 = 7.613) is similar to observed bioactivity values reported in the literature (pIC 50 = 7.823) [101] . This indicates that the generated MLR model is a good prediction model. Predicted activity values using the SVM model indicate a lower pIC 50 with regard to TamifluH. Nonetheless, both models indicate that all TCM de novo derivatives are good candidates with neuraminidase inhibitory activity. MLR and SVM models for predicting hemagglutinin inhibitory activity were not established due to the lack of available hemagglutinin inhibitor structures in the literature. Stability profile analysis. Root mean square deviation (RMSD) and total energy results from MD are summarized in Figure 5 and provide information on N1-ligand complex and ligand stability. During the 20 ns simulation process, the RMSDs of the four complexes ranged between 1.4-1.7 Å . Xylopine_2 stabilized after 15 ns, and the total energy of the complex equilibrated at 219,000 kcal/mol. The ligand RMSD of Rosmaricine_14 remained stable throughout the simulation, and no evident changes in total energy were observed after 17 ns. The RMSD and total energy of Rosmaricine_15 stabilized after 13 ns. Fluctuations in ligand RMSD was observed in TamifluH for the first 5 ns, but no changes were observed in ligand RMSD and total energy from 5 ns until the end of MD. The larger ligand RMSD fluctuations and higher total energy observed for Xylopine_2 may be attributed to its spatial structure and docking characteristics. Xylopine_2 consists of a bulky xylopine and a 2-aminopyridinium residue linked through 1 in a way similar to that of cis conformations ( Figure 6 ). It is well established that cisconformations are less stable than their transcounterparts, and thus may explain the higher total energy levels for Xylopine_2. In addition, Xylopine_2 binds to N1 though the aminopyridinium residue, allowing the xylopine structure more freedom to rotate and thereby increasing ligand RMSDs and total energy levels. H-Bond network during MD simulation. The H-bond occupancy of each compound in N1 is summarized in Table 3 Figure 7 . The distance between Xylopine_2 and Glu228 was maintained between 2-3 Å ( Figure 7A ). The distance of Rosmaricine_14 and Glu119 and Glu 228 also remained between 2-3 Å ( Figure 7B ). From Table 3 , Rosmarcine_15 and TamifluH did not form high occupancy H-bonds with Tyr402. Intriguingly, bond distance profiles indicate that Tyr402 was one of the key amino acids for H-bond formation ( Figure 7C , 7D). Tyr402 bond distance generally exceeded 2.5 Å in Rosmaricine_15, thus explaining the low occupancy rate in Table 3 . For TamifluH, the distance in the first ns was between 3-4 Å and then decreased to 2-3 Å from 1-20 ns, thus accounting for the low occupancy rate as well. Despite the low occupancy rates, the bond distance profiles suggest that Tyr402 is an important N1 binding site for Rosmaricine_15 and TamifluH. Possible mechanism for protein-ligand interaction. Insights to how ligand stabilization occurs within the protein binding site can be discerned from MD simulation. The H-bond formations with Leu224 at 2 and Glu228 at 3 and 4 ''sandwich'' the aminopyridinium and anchors Xylopine_2 ( Figure 6A ). However, the xylopine moiety remained unattached, causing strain to the compound and possibly contributing to large H-bond distance fluctuations ( Figure 7A ). At approximately 15 ns, attraction between Glu228 and 3 causes the terminal amine residue to torque towards Glu228 increasing the distance from Leu224 ( Figure 6A ). As a result, the stable H-bond with Leu224 was lost, and an additional H-bond with Glu228 was formed from 2. At the end of MD simulation, the primary binding force for Xylopine_2 was H-bonds formed with Glu228. In contrast to Xylopine_2, multiple binding sites secured Rosmaricine_14 and 15 within the binding site as reflected by the small H-bond distance fluctuations compared to Xylopine_2 ( Figure 7B , 7C). Rosmaricine_14 bound to Glu119, Glu228, and Arg293 through 6, 7, and 8 respectively ( Figure 6B ). The multiple attachment points anchor both the aminopyridine moiety and the rosmaricine moiety, reducing strain on the molecular structure as reflected by the low total energy and bond distance fluctuations. The increase in bond distance from Arg293 in Figure 7B starting at 3 ns was due to Arg293 intermolecular Hbond formation between the O atom and NH 2 residue. Nonetheless, all H-bonds were maintained throughout the MD simulation, suggesting good stability of Rosmaricine_14. The mechanism for stability of Rosmaricine_15 is similar to that of Rosmaricine_14. In addition to H-bonds at 9 with Glu228 and 10 with Arg293 which are identical to Rosmaricine_14, H-bonds are formed at 11 with Glu228, 12 with Thr226, 13 with Asn262, and 14 with Tyr402 ( Figure 6C ). These additional anchor points stabilized residues that were available for rotation in Rosmaricine_14, further lowering the total energy of the compound ( Figure 5C ). Within the anchored ligand, twisting of 15 contributes to H-bond fluctuations at Glu228 such as Figure 6D ). The stability of Tamiflu as a result of these binding anchors is reflected in the low total energy profile ( Figure 5 ) and small H-bond distance fluctuations ( Figure 7D ). The ability of the de novo derivatives to form stable H1-ligand complexes has also been assessed [92] . All de novo derivatives were capable of forming H-bonds at Glu83 and Asp103, the key binding sites on H1. The torsion angles of flexible bonds in each candidate when in complex with H1 and N1 are summarized in Figure 8 . In Xylopine_2, all monitored bonds were stable in H1 except for e ( Figure 8A ). The fluctuations could be attributed to the attraction between the amine group H atoms and Asp 103. When bound to N1, b was the primary location for torque changes in Xylopine_2. The recorded torsion angle changes at b support our previous speculation that the unattached xylopine moiety is a key source of instability for Xylopine_2. Torsion angle fluctuations of Rosmar-icine_14 in both H1 and N1 were mainly due to rotations at g and j ( Figure 8B ). Such changes are expected as the H atoms on the amine group continuously rotate to form H-bonds with key amino acids. Bonds in Rosmaricine_15 ( Figure 8C ) exhibited similar characteristics to those in Rosmaricine_14. Rapid rotations at the amine groups l and o are visualized by the recorded angle trajectories. NAG ( Figure 8D ) and Tamiflu ( Figure 8E ) both have relatively stable intermolecular torsion changes. This indicates that the lower stability of NAG and Tamiflu in H1 and N1, respectively, are not due to instability of their ligand structures, but may be attributed to weaker or unstable ligand-protein affinities. Hydrophobic interactions. Hydrophobic interactions also played a role in stabilizing ligands within H1 ( Figure 9 ) and N1 ( Figure 10 ) binding sites during MD. Due to differences in ligand structure and binding conformation, amino acids with which hydrophobic interactions were formed differed. In H1, amino acids involved in hydrophobic interactions included Pro82, Asp103, Asn104, Cys107, Cys153, and Pro154. More stabilizing interactions including H-bonds and hydrophobic interactions were observed in the N1 binding site (Figure 10 ). For the TCM candidates and Tamiflu, the spatial distribution of H-bonds coupled and hydrophobic interactions limits the free movement of ligands within N1, thus increasing stability of the N1-ligand complex. To further investigate docking features, CoMFA and CoMSIA models were built and validated using 27 neuraminidase inhibitors listed in Table S2 . The PLS analyses results for CoMFA and CoMSIA models are shown in Table 4 . The CoMFA model was generated using both steric and electrostatic fields and yielded a (Table 5) . The validated CoMFA and CoMSIA maps were used to assess ligand bioactivity. Contour of the de novo compounds at 20 ns MD simulation to the relative spatial positions of CoMFA and CoMSIA feature maps are shown in Figure 11 . In Xylopine_2, Rosmaricine_14 and Rosmaricine_15, the H-bond between the 2-aminopyridinium group and Glu228 matched the electropositive group feature of the CoMFA model ( Figure 11A,11C,11E ) and the H-bond donor feature in CoMSIA model ( Figure 11B,11D,11F) . The hydrophobic benzene structures of Xylopine_2 matched the steric favoring region of the CoMFA map and the hydrophobic feature of the CoMSIA map. The carbonyl groups in Rosmaricine_14 and Rosmaricine_15 which formed H-bonds with Tyr402 satisfied the H-bond acceptor feature in the CoMSIA model. TamifluH also contours to both CoMFA and CoMSIA models. The 3-methoxypentane group close to Arg293 and Asn344 matched the steric favoring region of CoMFA ( Figure 11G ) and the hydrophobic feature of CoMSIA ( Figure 11H ). This residue has similar characteristics to the 2aminopyridinium group in the de novo derivatives. In addition, the N-methylacetamide group in TamifluH, which forms H-bond with Tyr402, is located near the H-bond donor feature in CoMSIA. Though all compounds contoured to the N1 inhibitor features identified by CoMFA and CoMSIA, a critical difference was observed between TamifluH and the TCM de novo derivatives. All compounds except TamifluH formed H-bonds at Glu228. As Glu228 is a primary binding site of N1 [99] , ability of the TCM de novo derivatives to maintain stable binding with Glu228 during MD simulation supports the potential of these compounds as drug alternatives to TamifluH. Due to the lack of reported H1 ligand bioactivities in the literature, direct assessment of bioactivity through construction of CoMFA and CoMSIA models was not possible. Alternatively, indirect support was provided by assessing the ability of de novo derivatives to maintain contour to the N1 CoMFA/CoMSIA maps while forming interactions at key residues in H1, Glu83 and Asp103 [92] . As illustrated in Figure 12 the TCM de novo derivatives docked into the H1 binding site and formed critical interactions at Glu83 and Asp103 without losing contour to the CoMFA and CoMSIA maps. These results suggest that not only were the TCM de novo derivatives capable of docking into both H1 and N1, but that biological activity was also predicted in both binding sites, thus it is possible to develop dual-targeting drugs from the selected de novo derivatives. Important features for potential H1 and N1 inhibitors are summarized in Figure 13 . For H1, a salt bridge with Glu83 and H-bond donor and/or electrostatic interactions with Asp103 are important characteristics that should be met. Potential inhibitors for N1 should have salt bridge and/or H-bond formation at Glu228 and interactions with Asp293. These features can be used to identify or design novel drugs for H1 and/or N1. In the case of the TCM de novo derivatives from this study, each compound could structurally fulfill the requirements of both H1 ( Figure 13A,13B,13C ) and N1 ( Figure 13D,13E,13F) binding sites, thus supporting their potential as dual-targeting compounds. In this research, we identified Xylopine_2, Rosmaricine_14, and Rosmaricine_15 as the top three de novo derivatives exhibiting binding affinity to H1 and N1. Addition of a pyridinum residue to the native structures of xylopine and rosmaricine contributes to bond formation at key residues in both H1 (Glu83, Asp103) and N1 (Glu228, Arg292). The de novo derivatives were predicted as active by the SVM and MLR models, and contoured well to the 3D-QSAR models. The TCM de novo derivatives were able to maintain contour while forming key binding interactions in H1, thus providing indirect support for bioactivity in H1. The results of this study indicate that the TCM de novo derivatives not only can bind to, but can also exhibit biological activities in both H1 and N1. Key binding locations of the de novo derivatives include Glu83 and Asp103 for H1, and Glu228 and Arg292 for N1. Mutations currently attributed to oseltamivir resistance are located at H275 and N295S of the NA [103] . Since the key binding locations of the TCM derivatives do not overlap with those causing oseltamivir resistance, derivatives will be able to bind to viruses that are currently resistant to TamifluH. In addition, the de novo derivatives do not bind to amino acids in H1 or N1 that are prone to mutation ( Table 6, Table 7 ) [40, 104] , thus would likely be able to exert activity across a range of mutant H1N1 viruses. Last but not the least, multiple bond formations observed in MD provide additional insurance against possible mutations at key binding residues. In the case of a single point mutation, the de novo compounds will remain bound to the H1 and N1 sites through another key residue, therefore resisting the development of drug resistance in the virus. Based on the results and observations of this study, the TCM de novo derivatives may be attractive compounds for designing novel dual-target inhibitors for H1 and N1. Virtual screening, de novo derivative generation, and molecular dynamics (MD) simulation were performed using Discovery Studio Client v2.5.0.9164 (DS2.5; Accelrys Inc., San Diego, CA). The two-dimensional and three-dimensional structures of TCM compounds were generated using ChemBioOffice 2008 (Perki-nElmer Inc., Cambridge, MA). Comparative molecular field analysis (CoMFA) and comparative molecular similarities indices analysis (CoMSIA) models were constructed using SYBYLß 8.3 package (Tripos Inc., St. Louis, MO). Compounds from the TCM Database@Taiwan were docked to H1 and N1 protein active sites reported in our previous study [91] . All procedures were completed under the forcefield of Chemistry at HARvard Molecular Mechanics (CHARMm) [105] . The virtual screening process was performed using LigandFit. The conformational search method was based on the Monte Carlo algorithm. Rigid body minimization following initial ligand placement was completed using Smart Minimizer. Scoring functions used by LigandFit were DockScore. TCM compounds that docked into both H1 and N1 proteins were selected and then ranked by the sum of their H1 and N1 DockScore. TamifluH was used as the control for N1, and its N1 docking score was set as the minimum requirement. The top TCM compounds that passed the filtering were selected for de novo evolution. In de novo evolution, TCM compounds were placed into the H1 and N1 protein binding sites described previously, and Ludifragments were attached to the native structure. The new derivatives were generated in full evolution mode. Derivatives from de novo evolution were subjected to additional screening through Lipinski's rule [106] to rule out orally unstable or pharmacologically inapplicable compounds. As de novo products generated for H1 and N1 proteins differed, all de novo products were re-docked to H1 and N1 proteins to assess binding affinity. De novo products that docked into both H1 and N1 proteins were selected and ranked by the sum of their respective H1 and N1 DockScore. The top ten compounds with the highest DockScore were selected for further structure-based analysis. The 27 neuraminidase inhibitors used, including 24 training set compounds and 3 test set compounds, were adapted from Zhang's study [102] . Compounds were drawn using ChemBioOffice 2008 (PerkinElmer Inc., Cambridge, MA) and modified to physiological ionization using the Prepare Ligand function in DS 2.5. Bioactivity values (IC 50 ) were also obtained from Zhang's study though the original sources were not clarified, and converted to pIC 50 (log(1/ IC 50 )). Molecular descriptors of the compounds were calculated using Calculate Molecular Properties in DS 2.5 and the GFA was used to select the best representative molecular descriptors [107] . Utilizing the best representative molecular descriptors identified through GFA, MLR and SVM models were constructed using MATLAB (The Mathworks Inc., Natick, MA) and LibSVM [108] , respectively, and used to predict the bioactivity of TCM de novo compounds. The MD simulation was performed using the Molecular Dynamics package of DS 2.5. The complexes were created with a 10 Å solvation shell of TIP3 water around the protein. Sodium cations were added to each system for neutralization. Minimization using Steepest Descent and Conjugate Gradient were performed at 500 cycles each. Each protein-ligand complex was gradually heated from 0K to 310K over 50 ps, followed by a 200 ps equilibration phase. The production stage was performed for 20 ns using NVT canonical ensemble and trajectory frames were saved every 20 ps. SHAKE algorithm was applied to immobilize all bonds involving hydrogen atoms throughout the MD simulation. Long-range electrostatics were treated with PME method. Time step was set to 2 fs for all MD stages. The temperature coupling decay time for the Berendsen thermal coupling method was 0.4 ps. Post processing of the trajectory was performed using Analyze Trajectory module. Torsion angles of each bond were also monitored through DS 2.5. LIGPLOT [109] was used to generate schematic diagrams of protein-ligand interactions for each candidate and control in H1 and N1. CoMFA and CoMSIA models were constructed through the partial least square (PLS) analysis using previously described neuraminidase inhibitors [102] . The optimal number of components was obtained from leave-one-out method to yield the highest r 2 and q 2 values in non-cross validation and cross-validation, respectively. Biological activities of the TCM de novo compounds were evaluated based on contour to the generated 3D-QSAR map. Table 6 . HA mutation points between the 1918 H1N1 and H1N1/09 viruses. Table 7 . NA mutation points between 1918 H1N1 and H1N1/09 viruses. Supporting Information

Synthesis of an antiviral drug precursor from chitin using a saprophyte as a whole-cell catalyst

BACKGROUND: Recent incidents, such as the SARS and influenza epidemics, have highlighted the need for readily available antiviral drugs. One important precursor currently used for the production of Relenza, an antiviral product from GlaxoSmithKline, is N-acetylneuraminic acid (NeuNAc). This substance has a considerably high market price despite efforts to develop cost-reducing (biotechnological) production processes. Hypocrea jecorina (Trichoderma reesei) is a saprophyte noted for its abundant secretion of hydrolytic enzymes and its potential to degrade chitin to its monomer N-acetylglucosamine (GlcNAc). Chitin is considered the second most abundant biomass available on earth and therefore an attractive raw material. RESULTS: In this study, we introduced two enzymes from bacterial origin into Hypocrea, which convert GlcNAc into NeuNAc via N-acetylmannosamine. This enabled the fungus to produce NeuNAc from the cheap starting material chitin in liquid culture. Furthermore, we expressed the two recombinant enzymes as GST-fusion proteins and developed an enzyme assay for monitoring their enzymatic functionality. Finally, we demonstrated that Hypocrea does not metabolize NeuNAc and that no NeuNAc-uptake by the fungus occurs, which are important prerequisites for a potential production strategy. CONCLUSIONS: This study is a proof of concept for the possibility to engineer in a filamentous fungus a bacterial enzyme cascade, which is fully functional. Furthermore, it provides the basis for the development of a process for NeuNAc production as well as a general prospective design for production processes that use saprophytes as whole-cell catalysts.

NeuNAc is the most prevalent exponent of sialic acids [1] . In mammals, sialic acids are usually found as terminal residues of glycol conjugates on the outermost cell surface. As a result of their location and their negative carboxylate functionality, sialic acids play important roles in mediating cellular recognition and adhesion processes [2] and in the infection cycles of severe viral diseases, such as influenza viruses A and B [3] . In these cases, de novo-synthesized viral particles attach to their respective sialic acids at the cell surface. Neuraminidase (sialidase) activity is needed for the propagation of the virus in the host. Consequently, sialic acid derivatives are successfully applied in the therapy of such virus-related diseases. One well-known product that functions as a neuraminidase inhibitor is Relenza. Its active pharmaceutical ingredient is Zanamivir, which is a direct derivative of the NeuNAc precursor [4] . Traditionally, NeuNAc is prepared through extraction from natural sources, such as bird nests, milk, or eggs [5] , through the hydrolysis of colominic acid (a homopolymer of NeuNAc) in a culture broth of Escherichia coli K1 [6] , or through chemical synthesis [7] . Methods for NeuNAc production have included a chemo-enzymatic process [8, 9] , a two-enzyme reaction process [10, 11] , a biotransformation process using E. coli [12] , and an E. coli whole-cell system [13] . However, the requirement for ATP or an excess of pyruvate and the subsequent expensive downstream processing has kept the costs of NeuNAc production considerably high (current market price is $100/g). Chitin is considered the second most abundant biomass available on earth [14] . The estimated annual biosynthesis of chitin is more than 10 11 tons in marine waters alone [15] . Unlike cellulose, the other dominant biopolymer, chitin can serve as a source for both carbon and nitrogen (C:N = 8:1) [16] . This property suggests that chitin is an optimal resource for the production of NeuNAc (C:N = 11:1) because no additional nitrogen would need to be applied as it would be if glucose or cellulose were used as raw material. Chitin is found in the exoskeletons of arthropods, such as crustaceans (including crab, lobster, and shrimp) and insects (including ants and beetles), the cell walls of fungi, the radula of mollusks, and the beaks of cephalopods (including squid and octopi). This polymer is composed of β-(1,4)-linked units of the amino sugar N-acetylglucosamine (GlcNAc) that is currently produced using hydrolysis of deproteinized and demineralized crustacean shells [17] . Chitinolytic enzymes from fungi of the genus Hypocrea have been extensively studied for decades [18] . More recently, the chitinolytic enzyme system of H. jecorina has been studied using genome-wide analysis [19, 20] . Unlike their bacterial counterparts (e.g., Serratia marcescens [21] ), Hypocrea chitinolytic preparations have a high ratio of exochitinase to endochitinase activity and almost exclusively release monomeric GlcNAc from chitin [22] , which is another advantageous aspect of chitin compared to cellulose. Nevertheless, this raw material has not been adequately used. Therefore, the basic premise of this study was to exploit the potential of a saprophytic fungus to degrade the cheap biowaste chitin to its monomer GlcNAc and to further metabolize this product to NeuNAc. Engineering a NeuNAc synthesis pathway into Hypocrea The biosynthesis of NeuNAc begins with the formation of N-acetylmannosamine (ManNAc) from GlcNAc or UDP-N-acetylglucosamine (UDP-GlcNAc). In mammals, ManNAc is then phosphorylated to give ManNAc-6phosphate (ManNAc-6P). The second step involves the condensation of either ManNAc (in bacteria) or Mac-NAc-6P (in mammals) with phosphoenolpyruvate (PEP) to give NeuNAc or NeuNAc-9P, respectively. In mammals, NeuNAc-9P is then dephosphorylated to generate NeuNAc (see Figure 1 ). Hypocrea naturally degrades chitin almost exclusively to GlcNAc [22] . Therefore, the challenge was to engineer a pathway to convert GlcNAc to NeuNAc via ManNAc, which would enable the use of Hypocrea as a whole-cell catalyst. Lee and coworkers found that whole-cell extracts of several photobacteria could convert GlcNAc to ManNAc [13] . Among them, Anabaena sp. CH1 exhibited the highest GlcNAc 2-epimerase activity; consequently, they cloned and characterized a gene encoding GlcNAc 2-epimerase from Anabaena sp. CH1 (E.C. 5.1.3.8), which was used in the present study as a Hypocrea codon-optimized gene. For the second step (the condensation of ManNAc to NeuNAc), the currently used enzyme-catalyzed processes use a lyase, which requires an excess of pyruvate. Use of this incurs high downstream processing costs. Therefore, we used the NeuNAc synthase (EC 2.5.1.56) from Campylobacter jejuni [23] in the Hyprocrea process. This enzymatic step entails the use of PEP instead of pyruvate, which in the intended in vivo process is supplied by the fungus, thereby leading to an irreversible and more efficient reaction towards NeuNAc [24] . Moreover, the need for an excess of pyruvate becomes obsolete, and the resulting downstream process is significantly simplified. Similar to the GlcNAc 2-epimerase, the coding sequence for the NeuNAc synthase was codon-optimized for the usage in Hypocrea. The synthetic pathway is presented in Figure 1 . The complete nucleotide sequences for both genes encoding the recombinant enzymes, tbage and tneub, are shown in additional file 1. As the ability of the fungus to metabolize NeuNAc is an important issue, a possible uptake of NeuNAc by H. jecorina was investigated. Therefore, the fungus was pre-grown on glycerol in liquid culture, and half of the mycelium was autoclaved and half of it was harvested. The dead and living mycelia were transferred to glycerol-containing medium to study growth conditions or to medium without a carbon source to study resting cell conditions. NeuNAc was added to both media, and cultures were incubated for 8 h. Supernatants from all conditions were analyzed after incubation for 0 and 8 h by HPLC after derivatization using 1,2-diamino-4,5-methylenedioxybenzene dihydrochloride (DMB). Similar amounts of NeuNAc were present under all conditions regardless of whether the fungus was alive or dead ( Figure 2a ). This result indicates that NeuNAc uptake does not occur in H. jecorina. As a positive control experiment we did a similar experiment but instead of NeuNAc, GlcNAc was added to the media. As can be inferred from Figure 2b GlcNAc was completely consumed after eight hours under both growth and resting cell conditions when the mycelium was viable. Recombinant Hypocrea strains were generated using protoplast transformation of H. jecorina QM9414. In the derived strains, the two Hypocrea codon-optimized genes (without GST-tag) were placed under the control of either the H. jecorina pyruvate kinase (pki) promoter, which is a strong constitutive promoter, or the H. jecorina xylanase 1 (xyn1) promoter, which is a strict shut-off system if an inducer (e.g. D-xylose) is missing. Such a system was used to avoid interference of the introduced recombinant pathway with cell wall biosynthesis and consequently, biomass formation. However, when comparing both promoter systems the strong pki promoter did not lead to decreased growth, diminished cell integrity or other adverse effects (data not shown). Therefore, we used strains in which both genes were under the control of the pki promoter for further studies as we observed a remarkably higher Neu-NAc formation. Transcriptional analysis of the recombinant H. jecorina strains was done by RT-qPCR to compare expression of both inserted genes using sar1 (SAR/ARF-type small GTPase) as a stable reference gene [25] . Furthermore, the copy numbers of both genes was measured by qPCR of genomic DNA using pki as a reference, which in the native H. jecorina genome is present as a single copy gene. Based on these analyses a strain (termed PEC/PSC1) was chosen for further investigations because it showed the highest equal expression of both inserted genes. This was confirmed by the finding that this strain bears two copies of each recombinant gene in the genome. These data were also supported using Southern blot analysis (data not shown). Both recombinant enzymes were heterologously expressed as glutathione S-transferase (GST) fusion proteins in E. coli; the affinity chromatography purified proteins were used to confirm that their enzymatic capability was not altered by the codon usage adaptation and to provide a positive control for the enzymatic assays later on. To determine if the recombinant enzymes were functional, both GST fusion proteins were used in an enzymatic assay. The presence of GlcNAc and the formation of the intermediate product ManNAc and the final product NeuNAc were monitored using HPLC-MS analysis and results are presented in Figure 3 . Application of the GST-fusion proteins of both enzymes in the in vitro assay led to the formation of ManNAc and NeuNAc demonstrating that the synthetic genes are expressed as functional proteins (Figure 3a1 and 3b1) . According to the GST-fusion proteins, cell-free extracts of the recombinant H. jecorina strain PEC/PSC1 were applied in the enzymatic assay. The formation of ManNAc ( Figure 3a2 ) and NeuNAc could be detected (Figure 3b2 ). This demonstrates that both enzymes are also fully functionally expressed in the recombinant H. jecorina strain PEC/PSC1. Neither ManNAc nor NeuNAc was detected using cell-free extracts from the parental strain in the assay (Figure 3a3 and 3b3) , indicating that these pathways are normally not active in Hypocrea. To investigate the stability of NeuNAc in cell-free extracts of the recombinant strain, according cell-free extracts obtained from the cultivation in a bioreactor on chitin (vide infra) were spiked with NeuNAc and incubated for 24 h. As a control, a heat-inactivated cell-free extract was similarly treated. Using HPLC analysis after derivatization with DMB, similar amounts of NeuNAc were detected in both extract preparations (Figure 3c) , suggesting that components of the cell-free extract do not actively degrade NeuNAc. In addition, a similar amount of NeuNAc was measured in a NeuNAc-spiked cell-free extract of the recombinant strain that was not incubated, assuming that the 24-h incubation period at 30°C did not decrease the NeuNAc levels. As a final control, a cell-free extract without NeuNAc was also analyzed after a 24-h incubation period and, as expected, showed a lower amount of NeuNAc, which could only have resulted from its formation during the cultivation on chitin. In summary, we did not observe degradation of NeuNAc by H. jecorina. These data suggest that NeuNAc is not metabolized by the recombinant Hypocrea strain. We next addressed whether the recombinant H. jecorina strain had the ability to produce NeuNAc in vivo. To test this, the strain was grown on GlcNAc in shake flasks and cultivated on colloidal chitin in a bioreactor. Data on the corresponding cultivation monitoring are provided in additional file 2. As a positive control, an enzyme assay using the GST fusion proteins was again performed and resulted in the detection of ManNAc using HPLC-MS analysis as shown in Figure 4a1 . Notably, the intermediate ManNAc was detected in the recombinant strain, regardless of the carbon source (Figure 4a2 und Figure 4a4 ), whereas the parental strain did not form ManNAc (Figure 4a3 ). In the parental strain, only the first metabolite, GlcNAc, was detected, and it was present because it was either directly used as a carbon source or formed by degradation of the biopolymer chitin due to the native chitinolytic activity of the fungus. The synthesis of the product NeuNAc was analyzed using HPLC-MS/MS analysis (Figure 4b) . As a positive control, the reaction products (ManNAc, NeuNAc) generated by the use of the GST fusion proteins in an enzymatic assay are shown (Figure 4b1) . Importantly, the recombinant H. jecorina strain formed NeuNAc using either carbon source, Figure 2 The analysis of possible metabolization of NeuNAc in H. jecorina. The parental strain was pre-cultured on glycerol, and half of the mycelia were autoclaved. Living (light grey) and dead (dark grey) mycelia were transferred to MA media containing glycerol or MA media that lacked a carbon source, and NeuNAc (a) or GlcNAc (b) (as positive control) was added to both media. Samples were collected at 0 and 8 h. For NeuNAc analysis supernatants were derivatized using DMB befor analyses using HPLC. The presented values are the means of two biological duplicates that were derivatized in duplicate. Error bars indicate the standard deviations. GlcNAc or chitin (Figure 4b2 and 4b4) , whereas in the parental strain no formation of NeuNAc was detected (Figure 4b3 ). This analysis allowed us to estimate that 13 μg NeuNAc per g mycelium (dry weight) was formed in the recombinant strain. Thus, on its own, this would not be a competitive production process, but it does demonstrate the possibility for engineering a saprophyte and using it as a whole-cell catalyst that expresses a bacterial enzyme cascade. This method has enormous potential considering its use of a cheap starting material and the relatively simple, inexpensive cultivation of a fungus. Taken together, we successfully engineered Hypocrea in a way that this fungus now produces NeuNAc from the biopolymer chitin by employing its natural saprophytic activity in combination with the introduction of a bacterial enzyme cascade. Because human society will face severe bottlenecks in the supply of energy and in obtaining certain raw materials in the upcoming years, we hope that this study will highlight the potential advantages of biopolymers, such as chitin, and stimulate their efficient usage. Furthermore, we anticipate that such strategies will support efforts to create sustainable production processes. The parental strain H. jecorina (T. reesei [26] ) QM9414 (ATCC 26921) was maintained on malt extract (MEX) agar. Mycelia for the enzymatic assay were cultivated in 3% (w/v) MEX medium using 10 8 conidia/L at 30°C. Cultivation of H. jecorina on colloidal chitin was performed in a bench top bioreactor (Bioengineering, Wald, Switzerland) as previously described [27] . Briefly, 500 mL Mandels-Andreotti (MA) [28] medium containing 1% (w/v) colloidal chitin [29] , 0.5% GlcNAc, and 0.1% (w/v) bacto peptone (Difco, Detroit, US) was inoculated with 10 8 conidia/L. Some drops glanapon (Becker, Wien, Austria) were added to the medium to avoid excessive foam formation. Cultivation was performed at 30°C temperature, pH 5, 0.3 vvm aeration rate, and 500 rpm agitation rate for 96 h. Each sample drawing was followed by a microscopic analysis for infection control. Culture supernatant and mycelia were separated by filtration through GF/F glass microfiber filters (Whatman, Brentford, UK). All strains (parental, recombinant) showed similar growth on rich media as well as MA medium. The synthetic gene tbage (for sequence see additional file 1) is based on the protein sequence of Anabaena sp. CH1 GlcNAc-2-epimerase (GenBank: ABG57042) and was reverse translated into a nucleotide sequence using the GeneOptimizer ® software (Geneart, Regensburg, Germany). The codon usage was optimized for H. jecorina (http://www.kazusa.or.jp/codon). The synthetic gene tneub (for sequence see additional file 1) was similarly obtained based on the protein sequence from Campylobacter jejuni NCTC11168 NeuNAc synthase (http://old.genedb.org/ genedb/cjejuni/index.jsp, Cj1141). The synthetic genes tbage and tneub were excised from the production plasmid using XbaI/NsiI digestion and inserted into pRLM ex 30 [30] to generate the plasmids pMS-PEC and pMS-PSC. For the construction of pGEX-epi and pGEX-syn, the oligonucleotides GEXfw and GEXrev (Table 1) were used to introduce an XbaI and NsiI site into plasmid pGEX4T-2 (GE Healthcare, Chalfont St Giles, UK), yielding pGEX-MS. tbage and tneub were inserted into pGEX-MS via XbaI/NsiI digestion to yield the plasmids pGEX-epi and pGEX-syn. The protoplast transformation of H. jecorina was performed as described previously [31] . The plasmid pHylox2 (2 μg) [32] , which confers hygromycin B resistance [30] , and 4 μg of each plasmid pMS-PEC and pMS-PSC were co-transformed into the fungal genome. Fungal genomic DNA was isolated as described previously [31] . Southern hybridization and detection were performed using the DIG High Prime DNA Labeling and Detection Starter Kit II following the manufacturer's instructions (Roche, Basel, Switzerland). RNA extraction, cDNA synthesis and qPCR analysis were performed as described elsewhere [25] . Primer sequences are given in Table 1 . GST fusion proteins of GlcNAc-2-epimerase and Neu-NAc synthase were generated using plasmids pGEX-epi and pGEX-syn in E. coli BL21 (DE3). Purification of the proteins was performed using GSTrap™FF (GE Healthcare) according to standard procedures. Harvested mycelia were ground into fine powder and resuspended in 0.1 M Bicine buffer (pH 8) containing protease inhibitors (2 μM leupeptin, 1 μM pepstatin A, and 10 μM PMSF) (0.3 g mycelia/mL). The suspension was sonicated using a Sonifier ® 250 Cell Disruptor (Branson, Danbury, US) (power 40%, duty cycle 50%, power 20 sec, 40 sec pause, 10 cycles). Insoluble compounds were separated using centrifugation (10 min, 13000 g, 4°C). Enzymatic analysis was performed according to a previously described modified protocol [33] . The assay was performed in a total volume of 100 μL containing 10 mM GlcNAc, 10 mM PEP, 12.5 mM MnCl 2 , 100 mM Bicine buffer (pH 8) and 40 μL cell-free extract. Reactions were incubated for 60 min at 37°C, terminated at 85°C for 10 min and analyzed using HPLC. As a positive control, 5 μL of both GST fusion proteins were applied in place of the cell-free extracts. The stability of NeuNAc in the cell-free extract was determined by adding NeuNAc (150 μM) and incubating for 24 h at 30°C. After derivatization with DMB [34] , the NeuNAc quantity was measured using HPLC. Harvested H. jecorina mycelia were ground into fine powder and resuspended in water (0.3 g mycelia/mL). The suspension was sonicated using a Sonifier ® 250 Cell Disruptor (Branson) (power 70%, duty cycle 50%, power for 1 min, 1 min pause, 3 cycles). Insoluble compounds were separated using centrifugation (10 min, 13000 g, 4°C), and the supernatant was analyzed using HPLC-MS/MS. H. jecorina mycelia were pre-grown on MA containing 1% glycerol, transferred to MA medium containing 1% glycerol or no carbon source, spiked with 30 μM Neu-NAc or GlcNAc, respectively, and incubated for 8 h at 30°C. Autoclaved mycelia served as a negative control. After derivatization with DMB [34] , the NeuNAc quantity was measured using HPLC. NeuNAc, ManNAc and GlcNAc formation was measured using LC-MS (IT-TOF-MS) (Shimadzu, Kyoto, Japan) with a Rezex™ RHM-Monosaccharide H + -column (8%, 300 × 7.8 mm) (Phenomenex, Torrance, USA). The mobile phase consisted of water with 0.1% (v/v) trifluoroacetic acid, the flow was 0.6 mL/min, the column temperature was 80°C, and the injected volume was 10 μL. MS detection was performed in ESI+ mode, covering a scan range of 60-600 amu. The retention times were determined using pure standard substances. The identity of NeuNAc was confirmed by both, chromatographic retention time and mass spectral signal, which are very well matched by authentic standards of NeuNAc. The better the mass accuracy obtained from exact mass determination by HR-MS, the lower is the number of possible isobaric candidates (e.g. [35] ). In this case the mass accuracy is better than 2 ppm, leading to the number of candidates reduced to less than 10, with an even further reduction in the number of potential candidates because the isotopic pattern is also taken into account (what the software of the used IT-TOF-MS instrument does automatically). DMB derivatives of NeuNAc were separated on a Kinetex RP C18 (Phenomenex) at 0.75 mL/min with a 40°C column temperature and a mobile phase of water: methanol:trifluoroacetic acid (74.25:25:0.75). A Shimadzu RF-20AXS fluorescence detector (excitation 373 nm, emission 448 nm) was used for detection. Additional file 1: Coding sequences of the synthetic genes tbage and tneub. Coding sequences of the synthetic genes tbage and tneub. The sequences are provided in FASTA format. The XbaI site is underlined, and the NsiI site is double-underlined. The start codon ATG and the stop codon TAA are presented in bold letters.

mRNA pseudoknot structures can act as ribosomal roadblocks

Several viruses utilize programmed ribosomal frameshifting mediated by mRNA pseudoknots in combination with a slippery sequence to produce a well defined stochiometric ratio of the upstream encoded to the downstream-encoded protein. A correlation between the mechanical strength of mRNA pseudoknots and frameshifting efficiency has previously been found; however, the physical mechanism behind frameshifting still remains to be fully understood. In this study, we utilized synthetic sequences predicted to form mRNA pseudoknot-like structures. Surprisingly, the structures predicted to be strongest lead only to limited frameshifting. Two-dimensional gel electrophoresis of pulse labelled proteins revealed that a significant fraction of the ribosomes were frameshifted but unable to pass the pseudoknot-like structures. Hence, pseudoknots can act as ribosomal roadblocks, prohibiting a significant fraction of the frameshifted ribosomes from reaching the downstream stop codon. The stronger the pseudoknot the larger the frameshifting efficiency and the larger its roadblocking effect. The maximal amount of full-length frameshifted product is produced from a structure where those two effects are balanced. Taking ribosomal roadblocking into account is a prerequisite for formulating correct frameshifting hypotheses.

The reading frame of the vast majority of mRNAs is determined by the start codon after which the downstream cistron is translated in the same frame. Maintenance of the reading frame occurs without further signals to the ribosome. However, examples of genes containing information for programmed frameshifts can be found in most organisms, or in some of their IS sequences, transposable elements, retroelement-derived sequences or viruses. The sequence-information needed for programmed ribosomal frameshift varies and both +1 and À1 frameshifts can be induced (1) (2) (3) . Here, we focus on the frameshifting signal found in several viruses (1) , including infectious bronchitis virus (IBV) and SARS-CoV. The signal leads to programmed ribosomal À1 frameshift, whereby multiple proteins are produced from a single polycistronic messenger RNA (mRNA) (4, 5) . The frameshift efficiency, i.e. the fraction of ribosomes, which change reading frame, is important to ensure a correct stoichiometric relationship between the different products of translation. It has been shown that altered frameshift efficiency has detrimental effects on the proliferation of HIV-I and the yeast L-A viruses (6, 7) . In order to induce À1 frameshift, these viruses rely on three physical features on the mRNA: a heptanucleotide sequence, a spacer and a downstream structure (8) . The heptanucleotide sequence, called the slippery sequence, is where the À1 frameshift occurs and typically has the following sequence: X XXY YYZ, where X, Y and Z denote nucleotide species and spaces indicate initial reading frame. The spacer is a stretch of 6-9 nt positioning the ribosome correctly at the slippery site when encountering the downstream structure. The downstream structure is most often found to be a pseudoknot. The pseudoknot structure probably functions as a physical barrier deforming upon approach of the translating ribosome (9) , thereby assisting the frameshifting process; however, geometry and surface charge of the structure may also play a role for the frameshifting (10) . In bacteria and yeast, programmed frameshift signals can have rather different elements, as, e.g. the upstream Shine-Dalgarno binding element in the autoregulatory RF2 gene frameshift site first described in Escherichia coli (11) or the different pattern of the +1 frameshift stimulating heptanucleotide sequences present in Saccharomyces Ty elements (2) . However, many frameshift signals deviate little from those described for the virus-derived system used here and many signals are of such general character that ribosomes from different kingdoms of life will respond to them by shifting frame (12) . This happens not always with the same efficiency as in the original organism (12, 13) and there are even examples found where a frameshift element can direct the ribosomes into À2 or +1 frameshift depending on the test organism (14) . Here, we challenged E. coli ribosomes by constructing artificial frameshifting signals containing pseudoknot-like structures with strong stems. Using a refined frameshift assay, involving two-dimensional (2D) gel electrophoresis of pulse labelled proteins, we show that a significant amount of frameshifted ribosomes permanently stall within the strongest pseudoknots which therefore efficiently act as roadblocks. The small ribosomal subunits have been shown to be sensitive towards mRNA secondary structure in the process of translation initiation and mRNA structures can exclude initiation both in eukaryotes during the scanning process (15) and in prokaryotes for binding between the mRNA and the 3 0 -end of 16S RNA (16) . The fully assembled and translating 70S or 80S ribosomes seem to be more robust. It is, however, broadly accepted that mRNA secondary structures can function as obstacles to translating ribosomes (17, 18) although examples exists of large secondary structures in mRNA that are translated without any ribosomal delay (19) . Nevertheless, there is compelling evidence from in vitro experiments showing that ribosomes may pause upstream to such structures, most pronounced if the structures form pseudoknots (20) (21) (22) . Possibly the lack of rotational freedom in the helix of stem 1, due to the pairing in stem 2, makes pseudoknot structures harder to 'unzip' by the ribosome than simple stem-loop structures (23) . This may explain why pseudoknots can pause ribosomes. Examples from nature show the existence of diverse peptide sequences, often present in regulatory circuits, which will stall ribosomes (24), but to our knowledge, a permanent halt of ribosomes caused by mRNA structures has not been shown previously. Recent single molecule investigations suggest that the mechanical strength of pseudoknots correlate with the ability of the pseudoknot to stimulate frameshift (25) (26) (27) , at least in a certain interval. However, the calculated Gibbs free energy does not always correlate with frameshift efficiency. Not only the strength of the stems, but also the interaction between the loop and the stems might be of importance for the ability to induce frameshift and for the overall mechanical strength and brittleness of the structure. If the pseudoknot becomes too strong the ribosome, frameshifted or not, might not be able to open it and continue translation, whereby the pseudoknot acts as a roadblock. Often in literature (25) (26) (27) (28) (29) (30) (31) frameshifting assays were performed on constructs exhibiting the common feature that the stop codon for the normal reading frame was located at the entrance of the pseudoknot (or inside the pseudoknot) and the stop codon for the successful À1 frameshift was located downstream of the pseudoknot. In most frameshifting assays, the amount of frameshifting is determined by quantifying the amount of full-length frameshifted versus non-frameshifted products. However, for this to be a correct measure, the frameshifted ribosome must continue translation through the pseudoknot and beyond to the À1 frameshifted stop codon. If the À1 frameshifted ribosome permanently stalls inside the pseudoknot, it would falsely be interpreted as if the ribosome did not frameshift. Therefore, there is a serious pitfall in the classical methods which renders the amount of frameshifted ribosomes to be non-correctly determined, i.e. be underestimated, potentially leading to false hypotheses regarding the physical mechanism of frameshifting. The observation that strong pseudoknot-like structures can stop translation lead to the hypothesis that the largest amount of frameshifted product will be produced if the pseudoknot is mechanically strong but without a significant roadblocking effect. Most likely, this is exactly the balance exhibited by naturally occurring viral pseudoknots. Escherichia coli strain MAS90 [E. coli K-12, recA1 D(pro-lac) thi ara F 0 : lacI q1 lacZ::Tn5 proAB + ]. Liquid cultures were grown in minimal MOPS media (32) using glycerol as carbon source. Cultures were incubated with shaking at 37 C for at least 10 generations in the log phase prior to being used in frameshift assays. Pseudoknots were designed using custom-made software, which ensued that the codon usage was appropriate for expression in E. coli and that the sequences were likely to fold into the correct structure as determined by pknotsRG (33) . Hence, the resulting sequences are artificial pseudoknot-like structures and there is always a risk that the structure does not fold as anticipated. The selected sequences were synthesized by GeneScript and were subsequently inserted into plasmid OFX302 [containing slippery sequence, spacer and pseudoknot (25) ] between HindIII and ApaI restriction sites. The in vivo frameshift assays were performed as described previously (25) . Briefly, 1 ml of an exponentially growing culture was induced with Isopropyl b-D-Thiogalactopyranoside (IPTG) to a final concentration of 1 mM at an optical density of 0.4-0.7 measured at 436 nm (OD 436 ). After induction for 15 min, the culture was pulse-labelled with $10 mCi L-[ 35 S]-methionine for 20 s and chased with 100 mg L-methionine for 2 min before being transferred to 25 ml of chloramphenicol (100 mg/ml) on ice. The cells were harvested by centrifugation and proteins were boiled in SDS buffer and separated by 9% SDS-PAGE. The gel was dried and placed on a phosphor imager screen (Molecular Dynamics) and left to expose for 1-3 days. Relative amount of protein of the relevant polypeptides was quantified using ImageQuant software and the frameshift efficiency (e) was determined as follows: where V FS is the relative radioactivity in the frameshift product, n met,FS is the number of methionines in the frameshift product, V STOP is the relative radioactivity in the in-frame stop product and n met,STOP is the number of methionines in the in-frame stop product. Two-dimensional SDS-gels were performed as described (34) with a few modifications (35) using samples from the frameshift assay described above. The frameshift efficiency was determined as described for the frameshift assay above, although polygonal shapes were used to encircle the polypeptides of interest and quantify the relative amount of radioactivity in them. Polypeptides originating from stalled ribosomes were found as radioactive polypeptides with appropriate isoelectric point and molecular weight appearing on gels when the translated transcript contained a pseudoknot. These polypeptides were absent when a transcript without a pseudoknot was translated. The weakest stalled protein spots were difficult to distinguish from spots originating from endogenous gene expression on these gels (compare to the 0 construct in Supplementary Figure S5 ) and their determination is connected with some uncertainty. The statistical analysis used to compare the stalling efficiency between pseudoknot 22/6a and 22/6b was an unpaired one-tailed Student's t-test with a significance level of 0.05. Total RNA was extracted from 1.5 ml culture samples by the 'Hot-phenol' extraction method and separated according to size by electrophoresis on 1.2% agarose, 6% formaldehyde gels in recirculating 1xMOPS buffer. Capillary blots were performed onto Hybond-N + (Perkin Elmer) membranes, and the RNA was crosslinked to the membrane by 0.12 J/cm2 UV light in a Stratalinker 1800. Riboprobes covering mRNA sequences as described in Figure 4 were made by T7 RNA polymerase transcripts from the pMAS39 'downstream' template (19) or from templates made by PCR where one primer included 'hanging out' T7 promoter sequences (gene10 and lacZ 5 0 probes). The riboprobes were synthesized in the presence of 32-P-UTP and the final specific activity was about 40 Ci/mmol of nucleotide. Hybridization and stripping of membranes were performed following standard protocols (Amersham, Hybond-N+ booklet, 2006). The membranes were wrapped in Saran wrap and placed on a phosphor imager screen (Molecular Dynamics) and left to expose over night. Signals were visualized using ImageQuant software. We created a series of plasmids containing different pseudoknots and where the in-frame stop codon was placed either immediately upstream ('Upstream stop') or $150 nt downstream ('Downstream stop') from the pseudoknot ( Figure 1A ). The 'Upstream stop' constructs had an in-frame stop codon in the spacer between the slippery sequence and the pseudoknot. This caused non-frameshifted ribosomes to produce a 28 kDa polypeptide (gene10 from phage T7) while ribosomes undergoing a À1 frameshift continued through the pseudoknot and into lacZ producing a 148 kDa fusion protein of the T7 gene10 and lacZ sequences. In the 'Downstream Stop' constructs we replaced the UAA stop codon immediately upstream from the pseudoknot with a lysine codon (AAA). This change caused non-frameshifting ribosomes to continue through the pseudoknot and terminate at a downstream UGA codon producing a 37 kDa polypeptide. The pseudoknot constructs based on the plasmid OFX302 (25) are detailed in Figure 1B . We systematically increased the length of stem 1 and in pseudoknot 22/6a through 22/6c, we exchanged GC with UA base pairs, thus, gradually decreasing the stability of stem 1. Often, the number of ribosomes which undergo À1 frameshift has been determined from constructs such as our 'Upstream stop' constructs, by separating radioactively labelled proteins by SDS-PAGE and quantifying the relative amount of protein in each of the two polypeptides (28 versus 148 kDa). Given the limited resolution of SDS-PAGE, it is, however, impossible to clearly differentiate between polypeptides produced by ribosomes that terminate at the in-frame UAA stop codon and ribosomes that undergo À1 frameshift but stall within the pseudoknot. In order to overcome this problem, we invoked 2D SDS-PAGE (34) whereby polypeptides were separated not only by molecular weight but also by their isoelectric point (pI). While polypeptides originating from ribosomes stalled in the pseudoknot varied only slightly in molecular weight, they varied significantly in their pI. Based on the 'Downstream Stop' construct, we calculated a theoretical 2D SDS-PAGE assay of a growing polypeptide as consecutive codons are translated (shown in Figure 2A ). At around 28 kDa, the trace splits into two, the triangles denote the non-frameshifted product and the circles denote the À1 frameshifted product. Red symbols denote codons inside the pseudoknot. Experimental data originating from the 'Downstream Stop' construct is shown in Figure 2B , the theoretically expected features are indeed present, e.g. both the non-frameshifted (DS-stop) and the À1 frameshifted (FS) products are visible. The heat shock proteins GroEL and DnaK serve as landmarks on the gel. Interestingly, a series of polypeptides originating from ribosomes stalled inside the pseudoknot appeared (inside dashed red line). For comparison, a standard SDS-PAGE of the same sample is shown in Figure 2C , here, the second level of information (isoelectric point) is lost and the relative blurry bands are difficult to interpret. The results shown in Figure 2 revealed that a 1D SDS-PAGE assay could not firmly identify polypeptides originating from a À1 frameshifted ribosome stalled in the pseudoknot from the non-frameshifted product in a 'Downstream Stop' construct. In order to quantify the amount of À1 frameshifted ribosomes stalled inside the pseudoknot, we performed a 2D SDS-PAGE separation of the radioactively labelled proteins originating from the 'Upstream Stop' construct (Supplementary Figures S4 and S5) , which is the type of construct most commonly used throughout literature. The advantage of a 2D-gel analysis is that all the unfinished protein chains with different lengths concentrate in a common spot when they have the same pI. This made it possible to identify randomly stalled translation products inside the pseudoknot sequence and we quantified the amount of radioactivity in all identified additional spots. This produced a conservative estimate of the amount of stalled translations. The result of quantifying the fraction of in vivo À1 frameshifted ribosomes, both those which made it all the way to the lacZ stop codon (gene10/lacZ fusion) and those which stalled inside the pseudoknot, is shown in Figure 3A . The hatched bars denote the À1 frameshift efficiency taking into account only the end product of À1 frameshift (148 kDa gene10/lacZ fusion). This frameshift efficiency was calculated as (intensity of FS product)/ (intensity of non-FS product+intensity of FS product). The filled bars denote the À1 frameshift efficiency when both the end product (148 kDa gene10/lacZ fusion) and the products originating from stalled ribosomes are taken into account. This frameshift efficiency was calculated as (intensity of FS product+intensity of stalled product)/(intensity of non-FS product+intensity of FS product+intensity of stalled product). In addition to the six artificial pseudoknot-like structures, we also analysed two earlier investigated pseudoknots PK400 and PK401 (25), with over-all A B Figure 1 . Frameshift assay and pseudoknot structures. (A) All plasmid constructs contain an IPTG inducible promoter in front of T7 gene10 (light grey), a complete frameshift signal, and lacZ (dark grey). The frame shift stimulating pseudoknot-like structure is inserted downstream of gene10. Immediately, downstream from the pseudoknot lacZ is inserted in the À1 reading frame relative to gene10. In the 'Upstream Stop' construct the non-frameshifting ribosomes will translate gene10 and terminate at a UAA stop codon in the spacer sequence and produce a 28 kDa polypeptide. Ribosomes undergoing À1 frameshift at the slippery sequence translate lacZ thus producing $148 kDa polypeptide. In the 'Downstream Stop' construct the UAA stop codon is replaced by an AAA lysine codon thus resulting in $37 kDa polypeptide being produced by non-frameshifting ribosomes which terminate at an UGA stop codon downstream from the pseudoknot. (B) Sequence and structure of the inserted pseudoknots, the slippery sequence and the spacer. In pseudoknot, 10/6, 22/6a, 22/6b and 22/6c the first base in loop 2 has been removed in order to maintain the downstream reading frame (underlined). The boxed insert in panel B shows the structure and sequence of previously described constructs (25) . structures more similar to naturally occurring pseudoknots (Figure 1 insert), inspired from structures in the infectious bronchitis virus (22, 28, 30) . The pseudoknot structures in this type of virus are selected for their effects on vertebrate ribosomes, but the stem1 length variations were found to yield approximately the same relative stimulatory effect in E. coli (25) and suggest that stem1 strength is equally important for stimulating bacterial ribosomes to frameshift. All pseudoknots investigated stalled some fraction of the frameshifted ribosomes, however, significantly more ribosomes stalled in the artificial pseudoknots than in those resembling naturally occurring pseudoknots (PK400 and PK401). To quantify the amount of ribosomes stalling within a pseudoknot in vivo we calculated the ratio of (stalled+non-stalled frameshifted ribosomes) to (non-stalled frameshifted ribosomes), the result is shown in Figure 3B . For the IBV inspired pseudoknots, this ratio was close to 1 signifying that essentially no ribosomes stalled. However, the ratio was significantly larger than 1 for the more artificial pseudoknots which acted as roadblocks for a large amount of frameshifted ribosomes. The length of stem 1 did not significantly influence on the amount of frameshifted or stalled ribosomes. Interestingly, within pseudoknots with the same overall structure (22/6a-c) 22/6a stalls a significantly higher fraction of frameshifted ribosomes than 22/6b (verified by Student's t-test, n = 4, a = 0.05, P = 0.012), which again stalls more than 22/6c. Hence, the ability to stall a ribosome correlated with the strength of the pseudoknot base pairs, the stronger the base pairs the more frameshifted ribosomes were stalled. Earlier studies have shown that the insertion of sequences able to form mRNA secondary structures into a gene may cause the RNA polymerase to stall or invoke a target for endonucleolytic attacks (19) . Therefore, in our analysis of mRNA pseudoknot-stalled ribosomes, it was important to verify that there was no significant population of mRNAs that ended within the pseudoknot structure. If such truncated transcripts were abundant, it would be difficult to distinguish between protein products from ribosomes stalled within the pseudoknot and protein products originating from ribosomes ending translation at 'non-stop' mRNAs having their 3 0 -ends within the pseudoknot sequence. In the latter case translation would be terminated by tmRNA trans-translation thus rendering the protein products unstable due to the tmRNA-encoded tag (36) . In the following subsections 'Identification of transcripts from the T7gene10-PK-lacZ gene fusions', 'Messenger RNA stability' and 'Coupling between translation and transcription is required for full-length transcripts', we will show that the observed proteins did indeed originate from stalled ribosomes and that they were not caused by other effects. Identification of transcripts from the T7gene10-PK-lacZ gene fusions. To identify the major class of transcripts from our pseudoknot containing constructs, we made a northern blot with RNA from all strains used to measure frameshift frequencies, which are those containing the upstream stop. We used three different probes hybridizing either upstream of the pseudoknot, immediately downstream of the pseudoknot or in the very end of the lacZ reading frame ( Figure 4A ). Figure 4B -D, there was an unspecific hybridization from all three probes to the 23S and 16S ribosomal RNAs. In E. coli, ribosomal RNA constitutes between 80% and 90% of total RNA depending on growth conditions and some cross-hybridization to these species is often seen in northern blots. Here, the uninduced culture in Figure 4B -D, lane '0 no IPTG', made it possible to estimate the unspecific probing to rRNA and the two bands were used as size markers on the blots. Following induction with IPTG, all strains showed increased hybridization above the 23S RNA band compared to the uninduced control with all three probes. The so-called 0 construct was described in reference (25) , and contains a slippery sequence and the UAA stop codon but no pseudoknot-like structure. In all strains, except the one with the 0 construct, there were a distinct band (Fl) representing the expected full-length transcript. The full-length transcript reached from transcription start to the stem-loop structure downstream of the 3 0 -end of the lacZ open reading frame ('hp' in Figure 4A ). This mRNA stem-loop structure has been shown to stabilize the lacZ transcript by reducing 3 0 -end exonucleolytic attacks (37) . The core plasmid contained no distinct transcription termination signal after the lacZ gene, and accordingly we found transcripts that exceeded far beyond the full-length Fl band ( Figure 4B-D) . In the beginning of lacZ, $200 nt into the open reading frame, there is a site, called 'pt' in Figure 4A , where the RNA polymerase is caused to terminate if there is inefficient translation initiation of the lacZ gene (38) . In the 0 construct there is no pseudoknot to stimulate frameshift at the slippery site. Therefore, virtually no ribosomes were expected to follow the RNA polymerase from gene 10 into the lacZ part of our gene fusion. As expected, Figure 4B and C, lane '0' shows a prominent band ('SP' for stop polymerase) corresponding in size and probe-ability to this premature termination product. Also, corresponding low amounts of high molecular weight transcripts are detected for this construct. All the other constructs shown in Figure 4 contained frameshift stimulating pseudoknots and a inspection of the northern blot showed that the 'SP' bands probed with both gene10 and lacZ5 0 sequences were present in sizes which correspond to the sizes of the pseudoknots inserted. Messenger RNA stability. The wild type lacZ mRNA half-life is close to the average mRNA half-life in E. coli (120 s) and transcription takes close to 80 s due to the length of the lacZ gene (three times longer than the average gene). Therefore, a northern blot of wild type lacZ mRNA under steady state transcription will always include a lot of unfinished native transcripts, as well as mRNAs under degradation. Here, our gene10-lacZ fusion was even longer and transcription should take $120 s. Accordingly, all induced strains included in Figure 4 show a distinct smear of mRNA fragments recognized by all three probes. In order to examine the half-life of our artificial transcripts, we made experiments where transcription from the P tac promotor was stopped due to removal of the inducer ( Figure 5 ). Two minutes after IPTG removal, any remaining smear should originate from mRNA degradation because most of the RNA polymerase should have reached the end of the gene fusion at this time. From the experiment, shown in Figure 5 , it is evident that both the 'Fl' and the 'SP' mRNA fragments had a half-life close to the average 2 min E. coli mRNA half-life. In addition, both the pseudoknot containing constructs (10/6 and 22/6a) revealed the existence of a short mRNA fragment that was recognized only by the gene10 probe but not the lacZ5 0 and 3 0 probes (indicated by 'asterisks' in Figure 5 ). This fragment includes the transcription start in the 5 0 -end and the pseudoknot in the 3 0 -end. We suggest that the pseudoknot acts as an exonuclease barrier like the natural stem-loop structure in the 3 0 -end of the wild type lacZ transcript (37) and thereby induces a degradation intermediate of a distinct length with increased half life compared to unstructured mRNA sequences like those from construct 0. Alternatively, but not mutually exclusive, a pseudoknot acts like a rho-independent termination signal to the RNA polymerase. However, the sequences were not followed by a row of uridine residues, which would be necessary to make a stem-loop structure into a functional transcription terminator. Coupling between translation and transcription is required for full-length transcripts. The final test of our model for the transcription pattern in our artificial gene fusion was to establish translational coupling beyond the slippery sequence and into the polar termination site (SP) in lacZ. By changing the upstream stop codon between the slippery site and the pseudoknot region into a sense codon ribosomes should, frameshifted or not, follow the RNA polymerase into the beginning of the lacZ sequence. The 22/6a and the 0 constructs are the two constructs with the lowest frequency of frameshifting. Therefore, they have the least ribosome traffic into the lacZ sequence. Alteration of the UAA stop codon into a lysine AAA codon in the spacer between the slippery sequence and the pseudoknot changed the pattern of transcripts immensely. These two downstream stop variants ('DS. stop' in Figure 6 ), which did not contain a stop codon upstream from the structure, expressed significantly more full-length ('Fl') transcript and only insignificant amounts of premature transcription stop fragment ('SP') compared to their sister constructs containing the UAA stop codon upstream from the structure ( Figure 6 ). Our control construct, PK401, which stimulated 14% frameshift, showed no premature transcription stop fragment ('SP') and therefore no change in transcription pattern was observed as a consequence of removing the upstream UAA stop codon ( Figure 6 ) thus confirming that the major effect causing the 'SP' fragment is polarity in the lacZ gene and not transcription termination caused by the pseudoknot sequences. Also, the very short band marked by 'asterisks' that appeared from the 22/6a construct was not present in the 'DS. stop' variant ( Figure 6 ). This exclude this mRNA fragment to be causal for the appearance of stalled protein products, because 22/6a ('DS. stop') is the construct that caused the highest frequency of stalling (compare Figure 2 and Supplementary Figure S5 ). Our conclusion is that the stable proteins observed from within the pseudoknot structures (Figure 2 , Supplementary Figure S1 , S2, S4 and S5) were products from stalled ribosomes. The stalling of the ribosomes was directly caused by the tertiary structure and not by some secondary effect, as, e.g. stop codon-less mRNA fragments ending within the structure sequences. The structures analysed in this study are artificial and were designed to fold into pseudoknot-like structures with a gradually increasing mechanical strength. The mechanical strength was adjusted by changing the base pairs of the two stems, which seems to be a reasonable way of crudely varying the mechanical strength, as the energy involved in base pairing is higher than the energies involved in, e.g. the electrostatic interaction of the loop with the stems. It is, however, likely that the loop-stem interaction, surface charges or other players than just mechanical strength influence frameshift stimulating effect of mRNA structures. As there is a consensus in recent literature that pseudoknot mechanical strength correlates with frameshifting efficiency (23) (24) (25) , it was intriguing that the amount of frameshifted product was reduced by the stronger pseudoknot 22/6a compared to the weaker 22/ 6b or c ( Figure 3A ). This proved to be caused by stalling of a significant amount of frameshifted ribosomes by the strong pseudoknots ( Figure 3B ). Future studies will show whether significant stalling can also be caused by naturally occurring pseudoknots. Quantitative northern blot analysis was used to examine whether the observed translation products ending within the pseudoknot structure arose from fragments of mRNA produced either by low RNA-polymerase processivity or specific endonucleolytic attacks by RNases at the pseudoknot sequences. No evidence was found of a specific population of transcripts that could explain the amounts of protein products attributed to originate from pseudoknot-stalled ribosomes. Also, our protein-stability assay showed that the translational products from the stalled ribosomes were stable for at least 80 min (Supplementary Figure S1) , thus indicating that the stalled ribosomes are not rescued by tmRNA and that the stalled proteins do not originate from truncated mRNA. We also checked whether the protein products from within the pseudoknot structure could arise from very slow rather than permanently stalled ribosomes. A pulse chase experiment (Supplementary Figure S2 ) revealed that within 16 min there was no sign of a redistribution of label between the stalled spots and the stop codon-terminated downstream stop product, thus proving the possibility of very slow ribosomes to be unlikely. It is possible that the newly discovered ribosome rescue factor, ArfA (39) could be active at pseudoknot-stalled ribosomes and that nascent proteins would be more stable than if saved by tmRNA. However, as can be seen in Supplementary Figure S3 , the growth of strains expressing pseudoknot 22/6a was severely affected by induction and showed a decrease in growth rate correlating to the amount of stall product observed. Because ribosomes are limiting in growing cells (40) , the sequestration of ribosomes by engagement in induced overexpression of a gene from a plasmid will often cause a strain to grow slower than the uninduced counterpart. The enhanced reduction in growth rate upon induction of 22/6a compared to the 0 construct (Supplementary Figure S3 ) could indicate that stalled ribosomes were not rescued at a sufficiently high rate and we suggest that either the ribosomal rescue systems were titrated by the large amount of mRNA induced from the plasmid alleles, or alternatively, that no rescue is possible for pseudoknot-stalled ribosomes. Our results are in agreement with the observation that the amount of protein produced from an mRNA can be reduced when a pseudoknot is located upstream (29) . Also, they provide a possible explanation for the reduction in frameshift efficiency observed by, e.g. Napthine et al. (30) when increasing the thermodynamic stability of stem 1 above a certain threshold. This apparent reduction in frameshift efficiency (observed by 1D SDS-PAGE) could be caused by the fact that a significant fraction of the 'frameshifted' ribosomes permanently stalled within the pseudoknot. We propose that pseudoknot induced frameshifting efficiency can be viewed as a balance between two competing effects (as visualized in Figure 7) , the mechanically stronger the pseudoknot, the larger the frameshifting efficiency (25) (26) (27) , however, the stronger the pseudoknot the larger the likelihood of stalling the frameshifted ribosome, thus preventing the translation of full-length frameshift product. Possibly, evolution optimized viral pseudoknots to balance these two effects. Hence, in measurements of frameshifting efficiency it is important to take into account the roadblocking effect of mRNA pseudoknots. Figure 7 . Model of frameshifting efficiency. Increasing the strength of a pseudoknot causes the pseudoknot to induce frameshifting at a higher frequency. However, the stronger the pseudoknot the larger the likelihood that it will act as a roadblock for the ribosome, reducing the amount of frameshifted product produced. The optimal frameshifting efficiency is achieved by balancing the two contributions.

Secretory phospholipase A2 pathway in various types of lung injury in neonates and infants: a multicentre translational study

BACKGROUND: Secretory phospholipase A2 (sPLA2) is a group of enzymes involved in lung tissue inflammation and surfactant catabolism. sPLA2 plays a role in adults affected by acute lung injury and seems a promising therapeutic target. Preliminary data allow foreseeing the importance of such enzyme in some critical respiratory diseases in neonates and infants, as well. Our study aim is to clarify the role of sPLA2 and its modulators in the pathogenesis and clinical severity of hyaline membrane disease, infection related respiratory failure, meconium aspiration syndrome and acute respiratory distress syndrome. sPLA2 genes will also be sequenced and possible genetic involvement will be analysed. METHODS/DESIGN: Multicentre, international, translational study, including several paediatric and neonatal intensive care units and one coordinating laboratory. Babies affected by the above mentioned conditions will be enrolled: broncho-alveolar lavage fluid, serum and whole blood will be obtained at definite time-points during the disease course. Several clinical, respiratory and outcome data will be recorded. Laboratory researchers who perform the bench part of the study will be blinded to the clinical data. DISCUSSION: This study, thanks to its multicenter design, will clarify the role(s) of sPLA2 and its pathway in these diseases: sPLA2 might be the crossroad between inflammation and surfactant dysfunction. This may represent a crucial target for new anti-inflammatory therapies but also a novel approach to protect surfactant or spare it, improving alveolar stability, lung mechanics and gas exchange.

Phospholipase A2 biology Phospholipases A2 are a widely distributed group of enzymes primarily implicated in the turnover of membrane phospholipids and lipid digestion. They are also crucial for the inflammation pathways, as they are the first step for the production of eicosanoids and other inflammatory mediators [1, 2] . Secretory phospholipase A2 (sPLA2) is the low molecular, well conserved and secreted form of the enzyme. It is excreted into the alveoli mainly by macrophages and mast cells [1] [2] [3] . sPLA2 has a dual role, as it contributes to the inflammation pathway and it is also the main enzyme involved in the catabolism of surfactant [1, 2, 4] This complex proteo-lipid mixture is essential for the alveolar opening and the maintenance of an adequate gas exchange. sPLA2 is well known to be involved in lung inflammation and surfactant degradation based on animal and human studies in adults [4] . Therefore, it is conceivable that sPLA2 through either its pro-inflammatory role or the surfactant catabolism, might be involved in the pathogenesis of several critical respiratory diseases. Basic data have shown that both sPLA2 activity and expression are regulated by many factors including steroids, Clara Cell Secretory Protein (CCSP), Tumor Necrosis Factor-α (TNFα), Surfactant protein A (SP-A) and certain surfactant phospholipids, Interleukine-1 (IL-1) and some other cytokines [4] . Imbalance in the sPLA2 pathway due to different production of its modulators may account for increased surfactant degradation or lung tissue inflammation. Schematic representation of sPLA2 pathway is presented in Figure 1 : the possible roles of the enzyme in the pathogenesis of acute respiratory distress syndrome (ARDS), infant respiratory distress syndrome (iRDS), broncho-pulmonary dysplasia (BPD), infection related respiratory failure (IRRF) and meconium aspiration syndrome (MAS) are also illustrated. Data about the sPLA2 Figure 1 Possible involvements of sPLA2 pathway in the pathophysiology of critical respiratory diseases in infants. Full lines with arrows and hatched lines with squares indicate stimulatory and inhibitory actions on the enzyme activity and expression, respectively. Bold arrows show the direct consequences of the enzymatic activity in different diseases. ARDS: acute respiratory distress syndrome; BPD: bronchopulmonary dysplasia; iRDS: infants' respiratory distress syndrome; IRRF: infection related respiratory failure; MAS: meconium aspiration syndrome; sPLA2: secretory phospholipase A2; CCSP: clara cell secretory protein; IL-1β: interleukine-1 β; SP-A: surfactant protein-A; TNFα: tumor necrosis factor-α; MV: mechanical ventilation. role in each of the above-mentioned diseases are described in the following paragraphs. A wide body of literature suggests a role for sPLA2 in the development of ARDS and acute lung injury (ALI), its milder form. sPLA2 interferes with the surfactant activity and so reduces compliance [4] [5] [6] . sPLA2 starts a vicious cycle in which it damages the surfactant; since some surfactant components have their own inhibitory effect on sPLA2 activity and expression [7, 8] , the sPLA2-induced surfactant damage reduces this inhibition and thus the enzyme is able to further catabolize the surfactant phospholipids [4] . Thus, sPLA2 facilitates the action of other injurious agents against the lung epithelium, leading to further surfactant damage, alveolar collapse and respiratory impairment [4] . This process is also linked to the lung tissue inflammation, since TNFα and some other pro-inflammatory cytokines are strong sPLA2 inductors throughout the regulation of NFkB nuclear transcription factor [4, 7, 8] . Moreover, sPLA2 itself starts the inflammatory cascade, since it is the first step in the biochemical pathway leading to the production of arachidonic acid derivatives [9] . Inflammation may further inactivate surfactant, contributing to the above-described vicious cycle [4] [5] [6] . sPLA2 activity is raised in broncho-alveolar lavage fluid (BALF) in animal models of ARDS and in adult patients and this correlates with the clinical severity and mortality [4, 5, 8, 10] . We recently found raised enzyme levels in post-neonatal ARDS, similarly to the adult findings [11] . Moreover, respiratory syncytial virus (RSV) infection seems to cause a more severe ARDS because of the sPLA2 overexpression, triggered by the RSV itself [12] . Consistently, transgenic animals defective for the CCSP gene, experience higher inflammation and mucous production, when infected by RSV [13] . In animal models, the administration of sPLA2 inhibitors reduced lung inflammation and improved both compliance and oxygenation, especially if the inhibitor is administered early during the injury development [8, 14] . Similarly, the inhibitor was able to reduce sPLA2 activity in BALF of patients with iRDS, IRRF, MAS and postneonatal ARDS [15] . The importance of surfactant is well known in neonatal critical care. An inadequate surfactant production is the pivotal cause of hyaline membrane disease, also called infant respiratory distress syndrome (iRDS), the most frequent respiratory disease of preterm infants [16] . Although exogenous surfactant administration is curative in many preterm infants, long-term respiratory sequels are still a significant problem in this population, with 20% of the surviving preterm babies affected by BPD [17] . Moreover, the tiniest babies born at the limit of viability often require multiple surfactant administrations. Many of these very preterm deliveries are associated with infections and chorioamnionitis [18] . In these cases, inflammation lengthens the lung injury, decreasing the usefulness of exogenous surfactant and damaging the lung tissue [18] . In such situation, sPLA2 is likely to play a crucial role: we found increased sPLA2 levels in babies with iRDS comparing to normal term neonates [19] . sPLA2 was found to increase foetal neutrophil migration and so to enhance lung tissue inflammation [20] . Babies with higher sPLA2 activity are likely to be the ones needing repeated surfactant administrations and they are at higher risk for chronic lung disease. Some authors recently tried to administer CCSP to preterm neonates. CCSP is a natural inhibitor of sPLA2 in the lung [21, 22] . This drug has been given endotracheally together with surfactant [21] achieving a significant reduction in lung tissue inflammation. Similar results in terms of inflammatory markers and lung function have been obtained in animal models of iRDS and MAS [23] [24] [25] . Other authors have proposed the same approach with endotracheally administered budesonide, vehicled by surfactant [26] . Budesonide inhibits sPLA2 and has been associated with a decrease in the release of sPLA2induced pro-inflammatory cytokines [27] . Surfactant is the cornerstone of iRDS therapy and the sPLA2 inhibition could theoretically protect it, reducing the need for repeated doses and improving the long-term respiratory outcome. During sepsis or pneumonia surfactant may be inadequately produced and recycled or it may be inactivated by lung tissue inflammation [28, 29] . In these cases surfactant therapy is often less useful and does not achieve the clinical improvement usually seen in iRDS [28, 29] . Mortality rate for such condition is still remarkable in term infants and even higher in preterm babies, who often experience sepsis or pneumonia as nosocomial infections acquired in the intensive care units [16] . sPLA2 is raised in BALF of neonates with IRRF [19] . This is consistent with animal and cellular studies showing that bacterial membrane lipopolysaccharide is a potent inductor of sPLA2 [4] . Nonetheless, no definite data are available about the role of sPLA2 and its pathway during IRRF. Pancreatic sPLA2 has been indicated as a main etiological agent of MAS, one of the worst form of neonatal lung injury, characterized by massive surfactant inactivation, lung tissue inflammation and airway obstruction [30] [31] [32] . Meconium carries high amounts of sPLA2 and bile acids that are likely to contribute to lung injury, increasing sPLA2 activity and causing further surfactant inactivation [33] . Moreover, not only the pancreatic sPLA2 but also the pulmonary isoforms of the enzyme may be involved in the syndrome, [34] as lung sPLA2 production may be boosted by the meconium-induced release of pro-inflammatory cytokines [35] [36] [37] and through a specific cross-talk between different enzyme isoforms [38] . Consistently, we have recently found raised levels of pulmonary sPLA2 in BALF of patients affected by MAS when compared to their own meconium and to control babies [39] . MAS still has a mortality rate of about 50% and sometimes requires invasive treatments as broncho-alveolar lavage using saline/surfactant solutions or extra-corporeal life support [40, 41] . sPLA2 is also involved in neonatal bile acids pneumonia, a more rare form of lung injury, [42] in which neonatal lungs are challenged with the bile acids coming from the maternal circulation when the mother is affected by obstetric cholestasis [43, 44] . In this condition, the neonatal lung may experience a sPLA2 overactivation [33] due to the bile acids coming from maternal circulation. This may lead to severe respiratory failure, since bile acids increase the sPLA2 activity enabling the presentation of the phospholipid substrate to the catalytic site of the enzyme [45] . It is known that some sPLA2 gene polymorphisms are associated with chronic obstructive pulmonary or coronary artery disease [46, 47] . Given the wide role of sPLA2 in many critical respiratory conditions, an individual predisposition due to different polymorphisms is likely to exist. Nevertheless, data about sPLA2 genetics, its association with respiratory failure and its clinical severity have not been published. Available data allow hypothesizing a role for sPLA2 or its modulators in the pathogenesis, in the clinical severity and in the development of complications of the above mentioned types of lung injury. Our aim is to clarify such role, in order to better understand whether or not sPLA2 therapeutic inhibition might be a helpful strategy. To do that, we are planning to: 1. Identify the exact subtype(s) of sPLA2 produced and secreted into the alveoli during post-neonatal ARDS-ALI, MAS, iRDS and IRRF. This is important to know because sPLA2 inhibitors may have a different specificity for the various enzyme subtypes. In animal models, distinct sPLA2 subtypes have been associated to lung dysfunction [1, [48] [49] [50] . 2. Study the main modulators of sPLA2 expression and activity (TNFα, CCSP, SP-A and IL1). This will allow identification of possible pathway imbalances and eventually new therapeutic targets. 3 . Clarify what happen to sPLA2 and its pathway when exogenous surfactant is administered, as it usually occurs to preterm neonates. This will allow to understand if there is a link between sPLA2 activity/overexpression, the repeated need for surfactant and BPD occurrence. 4. Clarify if there is a genetic predisposition due to different sPLA2 genes polymorphisms which could lead to more severe clinical pictures in iRDS, ARDS, MAS or IRRF or to a long term negative outcome. This is the first study aimed at investigating the whole sPLA2 pathway in the above-described types of lung injury. To date, no study has addressed the functioning of the whole sPLA2 pathway, including the role of genetics, pathway modulators and related exogenous therapies that may affect it. This is essential because the diseases in which sPLA2 is thought to be important are basically different and the enzyme could play a different role through different subtypes, with different influence of its modulators and various response to the exogenous surfactant administration. Moreover, gene polymorphisms may play a role affecting the enzyme activity and so the clinical picture. Furthermore, many respiratory diseases potentially caused or influenced by sPLA2 are typical of newborn infants (e. g.: MAS, iRDS) or are present both in adults and in children, but with different causes and characteristics (e.g.: ARDS) [51] . Thus, data coming from animal studies or from adult experience cannot be directly applied to children and a specific study is warranted. The data coming from the present project will be crucial for future studies targeted at developing an anti-sPLA2 therapeutics. A multicenter design has been previewed and the project will be coordinated at the Laboratory of Clinical Molecular Biology of the University Hospital "A. Gemelli", Catholic University of the Sacred Heart in Rome. A Study group on Secretory Phospholipase in Paediatrics (SSPP) has been arranged and project coordinators will be a clinical pathologist/biochemist (Prof. E. Capoluongo, Head of the Lab) and a paediatric intensivist/neonatologist (Dr. D. De Luca). SSPP will consist of two working groups for this project: a. Laboratory group. This consists of biochemists and biologists experts in several molecular biology techniques applied to BALF specimens. These investigators will remain blinded to the clinical data, which will be known only to the project coordinators. b. Clinical group. This will consists of all clinicians -neonatologists and/or pediatric intensivists -working in intensive care units (at the University Hospital "A.Gemelli" or in so called "Collaborating centers") where patients' enrolment, samples and data collection will be performed. Clinical investigators will meet a project coordinator regularly before the beginning of the study. This will happen by tele-conference or by visiting the collaborating centre. All investigators will remain in contact during the entire project by e-mail and/or tele-conference. One of the project coordinator will also give a 24h/7d availability by phone in case of urgent matters. The project is still open and other intensive care units are welcomed to participate as collaborating centres. Interested colleagues should contact the corresponding author to discuss the study feasibility (please see at the end of manuscript). To accomplish the study purposes, the work will be subdivided in two phases: 1) clinical phase; 2) bench phase. Enrolment The following group of patients will be identified: iRDS, IRRF, MAS, ARDS-ALI. To be enrolled in a group babies must fulfil all the following inclusion criteria: A. Preterm neonates (gestational age ≤ 37 sett) with iRDS C-Reactive protein (CRP) < 10 mg/L or procalcitonin (PCT) < 0.6 ng/mL in the first 72 hours of life; Chest-Xrays typical for iRDS; no clinical signs of sepsis; need for mechanical ventilation. B. Infants and neonates with IRRF (regardless of the age) B1. Early IRRF. Neonates from mother with vaginal or urine positive cultures. Respiratory distress signs and CRP > 10 mg/L [52] or PCT > 0.6 ng/mL [53] in the first 72 hours of life; clinical signs of sepsis or blood/BALF positive culture; need for mechanical ventilation. B2. Late IRRF. Neonates with respiratory distress signs beyond the first 72 hours of life or infants, irrespectively of the age and CRP > 10 mg/L [52] or PCT > 0.6 ng/mL [53] ; clinical signs of sepsis or blood/BALF positive culture; need for mechanical ventilation. C. Neonates with MAS Neonates with meconium stained and thick amniotic fluid who required broncho-aspiration following Neonatal Resuscitation Program guidelines [54] . Continuous need for mechanical ventilation at 15 minutes of life. Chest-X rays typical for MAS. D. Infants with ARDS-ALI [55] Infants beyond neonatal age (> 30 days of life) and < 1 year of age under mechanical ventilation and having PaO 2 /FiO 2 ratio < 200 (ARDS) or < 300 (ALI), chest-X rays typical for ARDS-ALI, acute onset of the respiratory distress and no cardiogenic oedema/increase in left atrial pressure. A control group has also been previewed, as follows: Patients ventilated for non-pulmonary reasons (e.g.: anaesthesia, central nervous system diseases), PaO 2 /FiO 2 ratio > 300 or FiO 2 = 0.21, negative CRP and PCT, normal chest-X rays and chest clinical examination. A careful revision of the clinical characteristics will be done for each patient at the moment of discharge (or death). This will be done in each centre to ensure the appropriateness of diagnosis and internal validity. Procedures to be performed in the intensive care units 1. Broncho-alveolar lavage. This procedure will be performed as soon as possible from the fulfilling of the enrolment criteria. In case of neonates, broncho-alveolar lavage will be performed in the following schedule: • PRE-SURFACTANT • POST-SURFACTANT (after at least 12 hours from the surfactant administration) • PRE-2 nd SURFACTANT (only for babies needing a second dose) • POST-SURFACTANT (after at least 12 hours from the 2 nd surfactant administration) Obviously, for infants receiving just a single surfactant dose or no surfactant at all, only one or two bronchoalveolar lavages will be carried out. This procedure is to be intended a non-bronchoscopic lavage: it will be performed according to our previously described and well standardized technique [15] and following the advices of the European Respiratory Society guidelines [56] . All BALF specimens will be added with 0.9% saline up to 2 mL and a small aliquot of the fluids will be sent for microbiological culture. 2. 1.5 mL blood drawing in a vial with no anti-coagulant to be centrifuged (see below). If a baby undergoes repeated broncho-alveolar lavages, the blood drawing will be repeated each time. Every BALF and blood specimens must be obtained within 1 hour from each other. 3. 0.5 mL blood drawing into an EDTA vial to be immediately stored at 4°C. This blood will be used for DNA extraction to analyse sPLA2 genes polymorphisms and will be drawn only once for each baby. In general, blood drawings will be performed from an indwelling arterial line or from a central venous line to avoid haemolysis. Blood without anti-coagulant and BALF samples will be immediately centrifuged at 3000 rpm for 10 minutes to separate the serum or the supernatants which will be immediately stored at -80°C. Data to be registered in the intensive care units The following data will be recorded either from the vital parameters monitors, from the ventilator screen or the clinical files. • Type of mechanical ventilation provided • Peak inspiratory pressure, positive end-expiratory pressure, mean airway pressure (Pāw), Expired tidal volume (pro Kg) • Total respiratory rate and spontaneous respiratory rate (if any) • Dynamic compliance over ten mechanical breaths or static compliance using end-expiratory occlusion (depending on the ventilator) § • Total respiratory system resistances over ten mechanical breaths (If patients are ventilated with high frequency oscillatory ventilation, instead of the above-mentioned parameters, Pāw, amplitude and frequency will be recorded. If a specific flow-sensor [57] is available the tidal volume delivered during oscillations will also be registered and used for further calculation [see below]). • Cumulative dose of exogenous surfactant (for neonates) • FiO 2 • Oxygen saturation at the right hand These data must be recorded as close as possible to the broncho-alveolar lavage/blood drawing (max within 1 hour from such procedures). These data will be recorded in real time in an appropriate electronic database provided by the coordinating centre to each intensive care unit participating in the study. Moreover, using the above-mentioned data, the following indexes will be calculated: • Oxygenation index (FiO 2 × Pāw/PaO 2 ) • PaO 2 /FiO 2 ratio • Ventilatory index (Peak -end expiratory pressure) × respiratory rate × PaCO 2 /1000 (if conventional ventilation is provided) • Alveolar ventilation estimate during high frequency oscillatory ventilation (DCO 2 = frequency * (tidal volume) 2 ], if a specific flow sensor is available) [58] Moreover, the following data will be recorded: • Mortality • Intensive care unit length of stay • Duration of invasive ventilation • Duration of oxygen therapy • Any neurological sequel at the discharge • Oxygen requirement after discharge • Diagnosis of chronic lung disease, for preterm neonates, according to the NICHD definition of BPD [59] . Exclusion criteria Patients with one of the following characteristics will not be enrolled in any group: 1. Congenital lung malformations of any type 2. Lung or thoracic surgery 3. Lung cancer of any type 4. Congenital complex malformations 5. Patients undergoing extracorporeal life support. Storage and transfer of data and samples All data will be anonymously stored in the above described electronic database and will remain property of the enrolling centre. At the end of the clinical phase they will be checked for validity in each centres and then sent in a secured way to the Coordinating centre. At that time all specimens will be also sent under dry ice to the Coordinating centre. 2) Bench phase sPLA2 pathway In serum and BALF supernatants the following analyses will be performed: • Western blotting for sPLA2-IIA, -V, -X. For this procedure external (actinin) and internal (recombinant human sPLA2 subtypes, -IIA, -V, -X) controls will be used and total protein measurements will also be performed with Bradford's method [5] . • TNFα assay • SP-A assay • IL1 assay These assays will be performed using specific ELISA/ EIA kits already used to analyse BALF. These methods have been proven to do not cross-react with other cytokines and with sPLA2; in previous studies, coefficients of variation of the standard curve resulted always ≤ 9% [15, 39, [60] [61] [62] . • sPLA2 global activity assay To do this assay, all samples will be centrifuged (for 10' at 12000 rpm and then for 3' at 3500 rpm) through a membrane-filter with a molecular weight cut-off of 30 kDa (Amicon Ultra centrifugal filter; Millipore, Billera, MA-USA), to separate the secretory and cytosolic phospholipases (which weight ≈14 kDa and ≈80 kDa, respectively) [39, 62, 63] . • High sensitivity urea nitrogen assay in the BALF supernatant. All measurements in BALF will be corrected for the serum-to-BALF urea ratio, as previously described [64] . sPLA2 genetics sPLA2-IIA [HGNC:9031], -V [HGNC:9038] and -X [HGNC:9029] genes polymorphisms will be studied in the patients' leukocytes. We found 15 single nucleotide polymorphisms (SNP) for these genes. These polymorphisms were searched in the dbSNP (http://www.ncbi.nlm.nih.gov/SNP/), JSNP (http://snp.ims.u-tokyo.ac.jp), GenBank at the NCBI (http://www.genbank.com) and Applied Biosystems genotyping databases (http://www.appliedbiosystems.com), as well as in a previous study linking them to coronary artery disease [47] . Analysis of genetic polymorphisms SNP genotyping will be performed by TaqMan allelic discrimination assay [65] . Polymerase chain reaction will be performed with specific primers at concentrations of 900 nM. Fluorescence data files from each plate were analyzed by a specific software. In order to verify the correct genotype assignment, we will randomly analyse some of the above screened samples by means of genetic sequencing (BigDye terminator technique). All laboratory procedures will be carried out respecting safety regulations and bench investigators will be blinded to the patients' group of origin and to their clinical data. To accomplish this blindness, before starting the bench phase all vials will be re-labelled with a new code and only the project coordinators will be aware of the new code. Statistics and sample size Data will be tested for normality and then analyzed with parametric or non-parametric procedures, as appropriate. Accordingly, univariate analysis using Student and analysis of variance or Mann-Whitney, Wilcoxon, Kruskal-Wallis and Friedman tests will be performed. Some laboratory data will be subjected to correlation analysis with clinical findings, using Pearson's, Spearman's or Kendall's technique, according to data characteristics. Subsequently, if needed, significant results will be subjected to multiple curve estimation procedure [66] and/ or multivariate analysis according to data characteristics and the results of the univariate analyses. The genetic data will be analyzed using χ 2 test for the Hardy-Weinberg equilibrium of alleles at the individual loci. The association between genotypes and clinical data will be tested with χ 2 -or Fisher test and then with logistic regression or analysis of co-variance, as appropriate [67] . All statistical analyses will be performed by the project coordinators, who have a long experience and formal training in biostatistics. Despite a formal sample size calculation is not warranted, based on the available data [5, 11, 19] , we previewed a convenience sample size, as follows: 50-60 preterm infants affected by iRDS, with at least 20 receiving a second surfactant dose; for ARDS and IRRF groups 20 patients will be also convenient, while 10 control neonates or infants will be considered. These sample size have been checked for power regarding the correlation between sPLA2 activity and two selected clinical variables (using Power and Precision demo rel. 3.2 [68] ). Given α-error of 0.05 and a correlation coefficient (r) ≥ 0.6 and ≥ 0.85 [5, 11, 19] for the respiratory compliance and the PaO 2 /FiO 2 ratio, respectively, the power resulted > 80% in both cases. The study is supposed to last 12-18 months for the enrolment phase in each collaborating centre, and 3-6 months for the laboratory phase at the coordinating centre. The protocol and consent form have been approved by the Ethical Committee of the University Hospital "A. Gemelli" at the Catholic University of the Sacred Heart (Rome, Italy) as coordinating centre. Local ethical boards in each collaborating centre have also approved the protocol. The participation to the study will not change in any way the routine clinical assistance previewed for every patient. Furthermore, the participation to the study will respect all the local Regulations about safety procedures and the privacy. Informed consent will be given by parents or tutors of each baby, before the enrolment in the study. Study results will be presented to each investigator by teleconference and/or e-mail. If possible a meeting in occasion of one of the major congresses in the field of Paediatrics or Critical Care (like the European Society for Paediatric Research or European Society for Paediatric and Neonatal Intensive Care Congresses) will be organised. Data will be also presented at these meetings and the subsequent manuscripts will be circulated between all investigators for revision. All resulting manuscripts will be authored by the project coordinators and by the group authorship (SSPP: Study group on Secretory Phospholipase in Paediatrics). This study, thanks to its multicenter design, will clarify the role(s) of sPLA2 and its pathway modulators in several paediatric and neonatal forms of lung injury. In fact, enough evidence is available to indicate sPLA2, or at least some of its subtypes, as a key point in the pathophysiology of certain critical respiratory diseases. Inflammation is a complex process and is essential in many of these conditions: sPLA2 might be the main crossroad between inflammation and surfactant catabolism. Since this latter is surely an important component of some clinical situations, sPLA2 is worth to be studied in its metabolic and genetic issues, trying to correlate them to the clinical pictures. Given the peculiarity of such diseases and the relative rarity of some of them, only a multicentre design will be able to clarify this field. The purpose is not free from practical consequences, as several sPLA2 inhibitors are now available or under advanced development [69] . Many drugs are already used in respiratory critical care [70] but sPLA2 blockade might represent a new antiinflammatory therapy and a novel approach to protect surfactant or spare it, improving alveolar stability, lung mechanics and gas exchange. The establishment of a net of centres involved in clinical research, along with the support of bench investigators, will help in understanding this field and in building future randomised interventional studies. Footnote § Compliance measurement depends on the type of ventilator. Those with the end-expiratory occlusion will provide a static measure, otherwise a dynamic measurement will be done using the hot-wire anemometer flow sensor. In that case, to increase the measurement accuracy, the spontaneous breathing must be temporarily avoided, gas leaks must be < 5% and stable respiratory conditions with minimal airway secretions must be achieved. Conditions and technique for this measurement have been described in details elsewhere [15] . The study has received funds by Catholic University of the Sacred Heart (part of Research fundings D1-2011, to E.Capoluongo) and by a charity program of a private engineering company (QProgetti srl, Rome, Italy, funds 2011), which has nothing to do with this research field and will have no role at all in the project.

Transmissibility and temporal changes of 2009 pH1N1 pandemic during summer and fall/winter waves

BACKGROUND: In order to compare the transmissibility of the 2009 pH1N1 pandemic during successive waves of infections in summer and fall/winter in the Northern Hemisphere, and to assess the temporal changes during the course of the outbreak in relation to the intervention measures implemented, we analyze the epidemiological patterns of the epidemic in Taiwan during July 2009-March 2010. METHODS: We utilize the multi-phase Richards model to fit the weekly cumulative pH1N1 epidemiological data (numbers of confirmed cases and hospitalizations) as well as the daily number of classes suspended under a unique "325" partial school closing policy in Taiwan, in order to pinpoint the turning points of the summer and fall/winter waves, and to estimate the reproduction numbers R for each wave. RESULTS: Our analysis indicates that the summer wave had slowed down by early September when schools reopened for fall. However, a second fall/winter wave began in late September, approximately 4 weeks after the school reopened, peaking at about 2-3 weeks after the start of the mass immunization campaign in November. R is estimated to be in the range of 1.04-1.27 for the first wave, and between 1.01-1.05 for the second wave. CONCLUSIONS: Transmissibility of the summer wave in Taiwan during July-early September, as measured by R, was lower than that of the earlier spring outbreak in North America and Europe, as well as that of the winter outbreak in Southern Hemisphere. Furthermore, transmissibility during fall/winter in Taiwan was noticeably lower than that of the summer, which is attributable to population-level immunity acquired from the earlier summer wave and also to the intervention measures that were implemented prior to and during the fall/winter wave.

Although the first known imported case of 2009 pandemic influenza (pH1N1) arrived in Taiwan on May 18 from the U.S. via Hong Kong, Serological evidence has indicated that the pH1N1 virus had spread to central Taiwan by April-June [1] . Local infections and laboratory-confirmed pH1N1 cases in Taiwan started to mount in significant numbers in July-August when the schools were in summer recess. By the time the schools reopened in September, multiple intervention measures had been implemented by the government, which include strict border temperature screening starting in May, a "325" class suspension policy [2, 3] implemented in September, and later a mass immunization program [3] [4] [5] starting in November. The number of cases began to decline by the end of the year, and continued to do so into early next year, until the government announced on February 23 the end of the fall/winter outbreak [6] with over 3000 laboratory-confirmed cases reported, 910 hospitalizations, and 41 deaths [7] . Although school closing was a widely used method of intervention around the world during the pH1N1 outbreak (see, e.g., [8] [9] [10] [11] [12] [13] ), its suitability, timing, and the manner of implementation remains controversial. When K-12 schools (kindergarten through high schools) reopened on August 31 in Taiwan, the government implemented a unique partial school closing policy called the "325" class suspension policy aimed toward kindergarten through secondary schools (K-9), cram schools, and after-school institutions. Under this policy, if within any three (3) consecutive school days, two (2) or more students in the same class are diagnosed with influenza, then that class will be suspended for the next five (5) days including weekends and holidays [2, 3] . The policy was designed to minimize the potential social impact of full-scale school closings in the event of a major influenza outbreak in the community; to detect cluster infections in school settings early and swiftly; and to contain the infections locally without disruption for the other students in the school. At the height of the class suspensions in late November, more than 1800 classes with more than 50,000 students from almost 800 schools in Taiwan were suspended on a single school day (Figure 1 ), yet without any visible disruption in the normal functioning of the society. Moreover, starting November 1, a mass immunization program was initiated in Taiwan sequentially, according to a priority list of 12 target groups [4] , with healthcare and public health personnel having the highest priority [5] . Subsequently, preschool children were immunized starting on November 9; and followed by pregnant women, K-6 schoolchildren, and people with major illness/injury being vaccinated starting on November 16; 7-12 year-olds on November 23; and the general population on December 12. By March 16, a total of 5.66 million doses of AdimFlu-S (unadjuvanted H1N1v from Adimmune) or Focetria ® (MF59 ® adjuvanted H1N1v from Novartis) were administered, and more than 5 million of the 23 million Taiwanese had been immunized [14] . Children 12 and under were advised to receive two doses of vaccine, although many of them eventually received only one dose due to various reasons. A simple mathematical model, the Richards model, is utilized to fit publicly accessible cumulative epidemic data in order to obtain estimates for the turning points (the peaks and volleys of the incidence curve) and the reproduction number R of a particular wave of infections. Examples of applications of the Richards model to infectious diseases include those of SARS [15, 16] , dengue [17, 18] , and the 2009 pH1N1 epidemic [19, 20] . In this study, we will make use of the Richards model to pinpoint the turning points of each wave of the epidemic, in order to ascertain the temporal changes of the epidemic in Taiwan in the summer months and during the fall and winter days. The transmissibility of the pH1N1 virus during the outbreak is determined through its reproduction number. The data was accessed from the Central Epidemic Command Center website of the Taiwan Centers for Disease Control (TCDC). Samples were collected from hospitals and clinics participating in the Taiwan Influenza surveillance system under the Taiwan National Influenza Center (Taiwan NIC), which was established in 2006 to integrate all existing efforts of influenza surveillance and notification with laboratory analysis systems throughout Taiwan in order to enhance the epidemic data collection capacity in Taiwan [21] . The weekly laboratory confirmed pH1N1 case data (by the week when the samples were collected and sent to the TCDC-contracted laboratories) and the weekly hospitalization data (by the week the lab-confirmed cases were hospitalized) from June 28, 2009 (epidemiological week or e-week 27 of 2009) to March 27, 2010 (e-week 12 of 2010) was accessed from the weekly Influenza Express made publicly available on the internet by the TCDC [22] during the epidemic. The surveillance protocols in Taiwan remained essentially the same throughout the data period since, by the time the data were collected, clinical characteristics of the pH1N1 infection had already been well understood from the spring outbreaks around the world. We also accessed the daily record of numbers of classes suspended and number of schools with at least one class suspended during the fall school semester (September 9, 2009 to January 20, 2010) from the TCDC daily pH1N1 updates [23] during the epidemic. The time series of class suspension data is given in Figure 1. Since this data is for school days only, the days are specified in the horizontal axis of Figure 1 in weekly increments of 5 school days, except for weeks with less than 5 school days at the beginning and the end of the school semester as well as the week containing the New Year holiday (January 1). The Richards model [24] is of the form: where the prime symbol "'" denotes the rate of change over time which is in eweeks. C(t) is the cumulative number of cases at time t (in weeks), K is the cumulative case number over a single wave or phase of outbreak, r is the per capita growth rate of the infected population, and a is the exponent of deviation. The explicit solution of the equation is Here the parameter t m is related to the turning point t i of a wave (or the inflection point of the cumulative case curve) by the simple formula t m = t i + lna/(ra), where ln denotes the natural logarithm function. Moreover, R 0 = exp(rT) where T is the generation interval of the disease, or the average time interval from the onset of one infected person to the time when the onset of his or her contacts occurs. It has been shown mathematically [25] that, given the growth rate r, the expression R 0 = exp(rT) provides an upper bound of the basic reproduction number regardless of the distribution of the generation interval that is being used. In this work, we will use the term effective reproduction number R instead, due to the community-level immunity likely achieved by July and the interventions implemented during the two waves. The Richards model is a phenomenological model which can be used to describe the phenomenon of a biological growth (of cumulative number in this case) without requiring detailed information on the actual process of disease transmission. The basic premise of the Richards model is that the incidence curve of a single wave of infections contains a single peak of high incidence, resulting in an S-shaped cumulative epidemic curve and a single turning point (or peak incidence) of the outbreak. The turning point, defined as the point in time at which the rate of accumulation changes from increasing to decreasing, or vice versa in the event of a multi-wave outbreak, can be easily pinpointed by locating the inflection point of the cumulative case curve, i. e., the moment at which the trajectory begins to decline, as demonstrated in previous applications (see, e.g., [15] [16] [17] [18] [19] [20] . This quantity has important epidemiologic implications, indicating either the valley (i.e., moment of acceleration after deceleration) or peak (i.e., moment of deceleration after acceleration) of a disease incidence curve. Multi-wave outbreaks also can be modeled by using the multi-phase Richards model [16, 18] . Simultaneous estimates of the model parameters r, a, t i , and K, based on fitting the explicit solution of the Richards model for C(t) to the epidemic data used in the study, can be obtained easily and efficiently using any standard software with a nonlinear least-squares approximation tool, such as SAS or Matlab. The procedure for locating multiple turning points for multi-wave outbreak, which required the use of the multistage Richards model, is detailed in [16] and hence is omitted here. We first fit the weekly laboratory confirmed pH1N1 case data by sample receiving week in Taiwan Table 1 with the model fit shown in Figure 2 . The turning points for the two waves are estimated at 8.50 weeks after e-week 29 and 7.96 weeks e-week 39, respectively. Subsequently, the weeks in which the turning points for temporal changes in the weekly confirmed pH1N1 case number took place on e-week 36 (8/30-9/5) for the first wave with a 95% CI range of (36.62, 37.38), and on e-week 47 (11/15-11/21) with a 95% CI range of (46.42, 47.50) for the second wave. We note that the above results were obtained by rounding off the estimates to the next largest integer, e.g., e-week 27+8.50 = 35.50 and hence e-week 36 is the week during which the turning point for the first wave occurred, and similarly for the second wave. To compute the effective reproduction number R, we use the generation time T = 1.91 days (95% CI: 1.30-2.71) for the 2009 pH1N1 in Mexico estimated by Fraser et al. [26] . We note that the given CI's for R 0 reflect the uncertainty in the generation time T as well as in the uncertainty in the least-squared estimates for r, and does not reflect the error due to the model itself, which is always difficult to measure. We also fit the weekly confirmed pH1N1 hospitalization data by hospitalization week in Taiwan from eweek 29 (7/12-7/18) of 2009 to e-week 12 (3/21-3/27) of 2010 to the Richards model. The results are given in Table 2 . The data also fit a two-phase Richards model with the first wave spanning e-weeks 27-39 (7/12/09-9/ 26/09) of 2009 and the second wave from e-week 39 (9/ 20-9/26) of 2009 to e-week-12 (9/27/09-3/27/10) of 2010 ( Figure 3) . The turning points for the weekly confirmed pH1N1 hospitalizations occurred on e-week 37 (9/6-9/12) for the first wave with a 95% CI range of (36.12, 36.82) and on e-week 46 (11/8-11/14) with a 95% CI range of (44.64, 46.62) for the second wave, which were the same weeks as the case number data turning points. The estimate for R using an estimated generation time T for pH1N1 in Mexico [26] is again provided. To further analyze and compare our previous results, we also make use of the daily class suspension data in Taiwan from September 9, 2009 to January 20, 2010, which allows us to ascertain the temporal changes in this intervention measure during the time period. Since this dataset started near the end of the first wave, according to our previous results, only one wave was modeled via the Richards model. The estimation results for model fit using the daily class suspension number data as well as the daily number of schools with at least one class suspended are given in Table 3 The actual confirmed case number (approximated by K in our model) is 1742 during the first wave and 3238 for the two waves. point. A graphical illustration of the temporal timelines of the epidemic, as illustrated by the three model fits, is given in Figure 6 . Moreover, an illustrative comparison of the estimates for R as obtained by the model fits is also provided in Figure 7 . In both Figures 6 and 7 , the results from fitting the number of schools with class suspended are omitted for brevity, since they are similar to that of the fitting with class suspension data. Figure 2 Model fit for the 2-wave Richards model using weekly confirmed pH1N1 case data by sample receiving week in Taiwan. The dots are the real cumulative data, the blue curve denotes the first wave, and the red curve denotes the second wave. The arrows indicate the weeks in which turning points had occurred. The actual number of confirmed hospitalizations is 297 for the first wave and 910 for the two waves. The estimates for effective reproduction number R obtained from the confirmed case and hospitalization data are in good agreement, with R in the range of 1.04-1.27 for the first summer wave during July-September, and 1.01-1.05 for the second wave in fall/winter, using the generation time estimated by [26] for the spring outbreak in Mexico. Serological evidence has indicated that approximately one in every ten persons was infected with the 2009 pH1N1 virus in central Taiwan by April- June [1, 27] ; hence the estimates using data after July does not yield, and can reasonably be expected to be lower than, the more commonly known basic reproduction number R 0 . A recent modeling study [28] of the 2009 pH1N1 epidemic by geographic region in Mexico reveals a threewave pandemic, with an initial wave in April-May (Mexico City area), a second wave in June-July (southeastern states), and a geographically widespread third wave in August-December. The estimates for the regional reproduction numbers R were 1.8-2.1, 1.6-1.9, and 1.2-1.3 for the spring, summer, and fall waves, respectively. The second and third waves in Mexico occurred, respectively, one month earlier than the summer (July-early September) and fall/winter (late September-March 2010) waves in Taiwan under study here and exhibit similar decreasing trend, although with higher R. Transmissibility of the fist pH1N1 wave in Taiwan during the summer in July-September, as measured by R, was lower than that of the earlier spring outbreak in North America [20, 26, 29, 30] and Europe [31] , most likely, at least in part, due to decreased social contacts among the population triggered by public awareness of the earlier, well-publicized outbreaks in Mexico and North America as well as the subsequent preemptive government campaign to reduce transmissions. It was also lower than that of the winter outbreak in the Southern Hemisphere around the same time [19, 32, 33] , perhaps attributable to the fact that it was the winter influenza season in the Southern Hemisphere. Moreover, It is lower than the final size estimate of R 0 (1.87; 95% CI: 1.68-2.06) obtained from serological study of a cohort household population in central Taiwan during the same period of time [1] . However, we note that this disparity is reasonable since the serologic data used for this estimate accounts for the asymptomatic cases among the cohort group. The decreased transmissibility (smaller R) during fall/winter can be reasonably attributed to increased community-wide immunity from the first wave, and perhaps to the 325 class suspension policy initiated in early September before the start of the fall/winter wave. Significantly higher estimate of R (focused on schoolchildren) in the range of 2.0-2.6 was found for the initial pandemic wave in Japan [34] . Using updated epidemic data and an age-structured model, the same authors also estimated R for the subsequent community-wide wave in Japan in early summer to be much lower (1.21-1.35) [35] , although different population and modeling methodology also may have played a role in the decrease in R in subsequent waves. Similar decreases in estimates of reproduction number of 2009 H1N1 when more than one pandemic wave had occurred have been reported in many countries, including Mexico [28] , Argentina and Brazil [19] , Canada [20] , and Japan [34, 35] . Furthermore, these studies show that it is not uncommon for multiwave outbreaks to be more transmissible in a first wave but less widespread with a smaller number of infections (or perhaps limited to a small subpopulation as was in the case of pH1N1 in Japan), when compared to subsequent waves. Moreover, the second wave in Taiwan started shortly after the school opened in September, when additional infections occurring in school settings (as demonstrated by substantial number of class suspensions) contributed to a large number of cases, but perhaps with relatively less per contact transmissibility when compared to household contacts, as it has been reported that sitting next to a case or being the playmate of a case did not significantly increase the risk of H1N1 infection [36] . The estimates for R using laboratory-confirmed case data by sample receiving weeks are slightly lower than those obtained by using confirmed hospitalization data. Although both the confirmed case and hospitalization datasets identify week 39 as the cutoff week for the two waves, the estimates of turning points for each wave differ by about one week when using the two datasets. Since only the more severe confirmed cases were hospitalized, the individuals in the resulting hospitalization time series is a selected subset of those in the confirmed case time series. Subsequently, the temporal trends of the two time series might not be closely comparable. However, the cumulative curves in Figures 2, 3 , 4, 5 indicate some similarity in the temporal trends of the cumulative data, mainly in the form of the turning points. The reproduction numbers of the two datasets, on the other hand, are indeed comparable since they mostly are generated from the initial growth rates and hence less affected by any selection bias. The confirmed case data is generated by sampling week, which could be different from the week of symptom onset and hence pose a potential source of some bias in data. However, samples were typically taken when the physicians diagnosed and reported H1N1 cases. We refer to 2003 SARS outbreak in Taiwan, when it was estimated that the onset-to-diagnosis interval is 1.20 days for previously quarantined persons and 2.89 days for non-quarantined persons [37] . Given the similarity in symptoms of SARS and influenza as well as the heightened public awareness due to the world-wide alarm over the seriousness of the pH1N1 pandemic by September, it is more than likely that the time delay from symptom onset to diagnosis (and sample collection) of pH1N1 cases in Taiwan would be no more, if not less, than that of 2003 SARS. Moreover, one would expect that the lesson of SARS and the subsequent efforts by the government to educate has taught the general public in Taiwan to avoid delays in seeking medical care. Subsequently, this delay of one or two days in the weekly data can be expected to be most likely not significant. The use of hospitalization data is mainly for the purpose of estimation of reproduction number and comparison with the resulting estimates using the confirmed case data, which is not affected by this delay that might be present in both data. Estimates of R obtained by using other (larger) estimated generation time in literature result in larger values for R, but generally are well within the ranges of the other studies (see, e.g., [19, 20, 26, [29] [30] [31] [32] and Table 2 [33]) and hence is omitted for brevity. Note also that the formula for R used here yields an upper bound over all possible distributions for T given the growth rate r, and hence might result in an overestimate of its true value. In Taiwan, the fall session for kindergarten to high school started on August 31, while the universities started the fall semester two weeks later, around mid-September. Our analysis using the weekly confirmed case and confirmed hospitalization data shows that the initial summer wave of pH1N1 epidemic in Taiwan had peaked by e-week 36-37 (8/30-9/12), around the time schools from kindergarten to grade 12 reopened on August 31. However, a second fall/winter wave of cases started to emerge near the end of September around eweek 39 (9/27-10/3), approximately 4 weeks after the schools reopened, which did not reach its peak until mid-November (e-week 46-47 or 11/8-11/21) and lasted until the turn of the year. It is interesting to note that the state-specific fall pandemic waves in Mexico began 2-5 weeks after school reopened [28] , which is consistent with our results on the start of the fall wave in Taiwan. Note that both turning points of the two waves in Taiwan fell on neighboring week using either the lab-confirmed case or hospitalization data. This is reasonable since the hospitalization of confirmed cases and the time that the samples were received by laboratories are closely related, although not necessarily in any particular order. The class suspension data started on September 9 near the end of the first wave when the earliest class suspension occurred, according to our 2-wave fitting in Tables 1 and 2 , hence only one wave was modeled via the Richards model (Table 3) . Moreover, November 19 (95% CI: November 18-20) was determined to be the turning point for the daily class suspension data, while November 17 (95% CI: November 16-18) is the turning point for the daily number of schools with class suspended. Both days fall on e-week 47, which coincides with the week where the turning point had occurred as pinpointed by using the confirmed case data and one week after the turning point obtained by using the hospitalization data. It is reasonable to expect the class suspension to take place following the occurrence of case reporting and hospitalization. Moreover, the use of daily data allows a more precise estimation of the turning point. Also of interest is the possible impact of major intervention measures implemented by the Taiwan government during this time period, which including the aforementioned "325 class suspension" policy and the mass immunization program. The daily number of class suspensions started to increase in early September and continued until late November after the implementation of mass immunization campaign (Figure 1 ). In particular, the 325 policy, which was designed to minimize the potential social impact of full-scale school closings in the event of a major influenza outbreak in the community; deserve special attention to ascertain its actual effectiveness. In fact, the lower estimates of R for the second wave and for the school closings data might indeed be attributable to the possible effects of school closings after September. However, more detailed class suspension data as well as age-specific epidemic data is needed to further quantify the actual impact or effectiveness of this very unique approach of partial school closure and localized class suspensions on the infections in the school and in the community in a qualitative modeling analysis (see, e.g., [12, 13, 38] ). Using routine influenza surveillance data, we modeled the temporal changes of the two waves of pH1N1 epidemic in Taiwan in summer and in fall/winter. The mass H1N1 vaccination program was first initiated sequentially on November 1, where a typical delay of at least two weeks from immunization is needed for protection from the vaccine to take effect in human bodies. Our results suggest that the turning point for the second wave of infections in the fall had occurred around mid-November (e-week 46-47 or 11/8-11/21). Moreover, the class suspension data indicate that the number of class suspensions had peaked by November 20, less than three weeks after the start of mass immunization and most likely before the impact of mass immunizations started to become significant. However, the mass immunization, and perhaps the voluntarily decreased social contacts by the general public in response to the well-publicized mass immunization campaign by the government, could have contributed to the overall mitigation of the disease in the community, as indicated by the early saturation of the winter epidemic by early February. However, this cannot be modeled without detailed vaccination data. The Richards model considers only the cumulative infected population size with saturation in growth as the outbreak progresses, which can be caused by other factors such as implementation of control measures. Although data by reporting date is often and typically scrambled by artificial factors such as health system alertness, public response, and government responsiveness, the Richards model is able to capture the turning points of outbreaks because they are often results of these artificial factors. We note, however, that the skewness in an epidemic curve, as quantified by the exponent of deviation "a" in the Richards model which describes the curvature of a given cumulative case data, also could conceivably arise from various other intrinsic factors such as spatial heterogeneity and individual heterogeneity in contact (see [39] , pp. 281 for example) which is not captured by this simple model. This type of modeling, although somewhat simplistic and subsequently limited in its quantification of complex factors, nevertheless enables us to ascertain the impact of these artificial factors through the temporal changes of an outbreak, especially in the events when detailed epidemic data describing disease transmissions and other relevant data (such as that of intervention measures in this case) are not readily available for the construction of a complete disease transmission model and the reliable estimation of model parameters, as in this study Moreover, the use of cumulative numbers could often, or at least partially, smooth out stochastic variations that typically occur in epidemic data, and hence the Richards model could be a valuable tool in providing clues to the challenging task of public health policy evaluation and planning.

Molecular mechanisms of inflammation and tissue injury after major trauma-is complement the "bad guy"?

Trauma represents the leading cause of death among young people in industrialized countries. Recent clinical and experimental studies have brought increasing evidence for activation of the innate immune system in contributing to the pathogenesis of trauma-induced sequelae and adverse outcome. As the "first line of defense", the complement system represents a potent effector arm of innate immunity, and has been implicated in mediating the early posttraumatic inflammatory response. Despite its generic beneficial functions, including pathogen elimination and immediate response to danger signals, complement activation may exert detrimental effects after trauma, in terms of mounting an "innocent bystander" attack on host tissue. Posttraumatic ischemia/reperfusion injuries represent the classic entity of complement-mediated tissue damage, adding to the "antigenic load" by exacerbation of local and systemic inflammation and release of toxic mediators. These pathophysiological sequelae have been shown to sustain the systemic inflammatory response syndrome after major trauma, and can ultimately contribute to remote organ injury and death. Numerous experimental models have been designed in recent years with the aim of mimicking the inflammatory reaction after trauma and to allow the testing of new pharmacological approaches, including the emergent concept of site-targeted complement inhibition. The present review provides an overview on the current understanding of the cellular and molecular mechanisms of complement activation after major trauma, with an emphasis of emerging therapeutic concepts which may provide the rationale for a "bench-to-bedside" approach in the design of future pharmacological strategies.

Despite significant advances in injury prevention, prehospital resuscitation strategies, and modern intensive care, trauma remains the main cause of death in young people in the United States, resulting in more years of potential life lost before the age of 75 years than any other disease [1] [2] [3] [4] . Until present, the pathophysiology of major trauma remains poorly understood [5, 6] . In principle, the pathophysiological sequelae of major injuries are characterized by the initial traumatic impact (so-called "first hit"), followed by a cascade of subsequent immunological reactions, which render the patient susceptible to a potentially detrimental "second hit" insult [7] . The activation of innate immune response mechanisms has been characterized as a crucial event initiating the early phase of hyperinflammation within hours to days after major trauma [6] [7] [8] . While innate immunity is classically considered to be the immediate "first line of defense" against non-self antigens (e.g. infectious pathogens), a traumatic insult can induce a similarly potent acute inflammatory response [9] [10] [11] [12] [13] . The trauma-induced immune response may be limited locally, as in isolated injuries, or result in a massive systemic immune activation, as in patients with multiple injuries [1] . The endogenous triggers of trauma-associated inflammation have been thoroughly investigated and characterized in recent years [7, 14] . The so-called "first hit" induced by a traumatic impact leads to the appearance of an arsenal of "damage-associated molecular patterns" (DAMPs) that are recognized by receptors of immune cells [15] . DAMPs represent a recently characterized large superfamily of danger signals which can activate innate immune responses after trauma or trauma-induced complications, such as infection and sepsis [7, 16] . The DAMP family of danger signals includes the so-called "pathogenassociated molecular patterns" (PAMPs) and molecules termed "alarmins" [17] . The list of molecules belonging to the DAMP family has been increasing dramatically in recent years, and their pathophysiological function in mediating trauma-induced inflammation is far from being fully understood [18] . PAMPs represent a heterogenic entity of recently described inflammatory molecules related to the innate immune system [17, 19] . These microbial molecules are recognized by the immune system as foreign due to their characteristic molecular patterns. In contrast, the so-called "alarmins" represent the correlate of PAMPs for all non-pathogen-derived danger signals which originate from tissue injury [17] . This heterogeneic group of danger molecules is capable of activating innate immune responses in response to tissue damage and cell injury. The alarmins comprise the "heat-shock proteins" (HSPs), annexins, defensins, as well as "classical" markers of tissue injury, such as the S100 protein and the high mobility group box 1 (HMGB1) protein [17, 20] . Immunologically competent cells recognize both PAMPs and DAMPs through multiligand receptors expressed on their surfaces, such as Toll-like receptors (TLRs) [21, 22] . The very early stage after tissue trauma is characterized by activation of cellular and molecular effectors of the innate immune system, including complement activation and recruitment and activation of neutrophils (polymorphonuclear leukocytes; PMNL) [6, 7] . The complement system appears to represent the crucial effector of innate immune responses in the early phase after major trauma [23] [24] [25] . Once the cascade is activated through one of three (five) established pathways (Figure 1 ), complement plays a critical role in the elimination of invading pathogens by opsonization for phagocytosis (C3b, C4b), chemotaxis of leukocytes (C3a, C5a), and by direct lysis of pathogens through the membrane attack complex (MAC, C5b-9) [23, 26, 27] . The generation of anaphylatoxins C3a and C5a provides potent chemoattractants for phagocytes and neutrophils, and recruit these immune cells to the site of injury [24, 28, 29] . The anaphylatoxins further induce degranulation of mast cells, basophils and eosinophils and mediate the hepatic acute-phase response [30, 31] . Finally, the generation of C5b by cleavage of C5 initiates the terminal complement pathway with MAC formation. The MAC forms through the self-association of C5b along with C6 through C9 and leads to the formation of a large membranolytic complex capable of lysing prokaryotic and eukaryotic cells [32] . Multiple previous studies have unequivocally shown that trauma activates complement, both locally at the site of injury, and systemically. Early studies in the 1980s revealed that the complement cascade is activated at the level of C3 in serum of trauma patients, and the extent of activation correlates with the severity of injury [33, 34] . The neutrophil (or PMNL) has been established as the cellular counterpart to the humoral immune response mediated by complement activation, and represents a "key effector" cell of the early posttraumatic immune response. Within minutes, and up to several days after injury, neutrophils play an important role in mounting the immunological defense and the debridement of injured tissue. Primed neutrophils are capable of mediating an inflammatory response, characterized by release of cytokines, chemokines, reactive oxygen species, and tissue-toxic enzymes, such as myeloperoxidase and elastase [20, 35] . Aside from the beneficial role of neutrophils in host-defense and clearance of damaged tissue after trauma, excessive priming and cellular PMNL activation may lead to an overwhelming inflammatory response and "innocent bystander" injury to host tissue [35, 36] . Uninjured tissue may become damaged by the local release of toxic metabolites and enzymes, thus contributing to remote organ injury (e.g. to brain and lungs), by contributing to tissue edema and secondary tissue damage [12, 35, [37] [38] [39] . Based on the delicate balance between protection and harm, the posttraumatic inflammatory response has been rightfully termed a "double-edged sword" [40] [41] [42] . The present review will outline the current understanding of complement activation and regulation after major trauma, with a focus on specific injury patterns, including musculoskeletal trauma, ischemia/reperfusion, chest and brain injuries. We will furthermore discuss potential new pharmacological strategies related to the targeted inhibition of complement, which may shed some hope into the design of new immunomodulatory treatment modalities for severely injured patients in the future. The complement system represents one of the phylogenetically oldest cascade systems of the body, consisting of a proteolytic cascade of more than 30 soluble and surface-bound proteins that can be activated by the classical, the lectin and the alternative pathway [32, 43, 44] . Recently, two additional complement activation pathways have been described, i.e. the properdin and the thrombin pathways, both of which will be discussed in more detail below. Figure 1 depicts a rough schematic of the so far known complement activation pathways and of the biological functions of activated complement components. In brief, the three main activation pathways converge in the formation of enzymatic complexes termed the C3 convertases and C5 convertases, which cleave the two main components of the complement system, C3 and C5. The two proteolytic fragments generated by the action of the convertases are the anaphylatoxins C3a and C5a. Both can trigger proinflammatory signaling through binding to their corresponding receptors, the C3a receptor (C3aR) and C5a receptor (C5aR and C5L2), on various myeloid and non-myeloid cells [28, 29, 45, 46] . C5a is a powerful chemoattractant for neutrophils that recruits immune cells to the site of injury and activates cellular attack mechanisms like oxidative burst and lysosomal enzyme release [47, 48] . Furthermore, the anaphylatoxins contribute to the degranulation of mast cells and basophils, induce the expression of adhesion molecules on endothelial cells, cause smooth-muscle contraction and enhance the acute phase response of the liver [48] . The cleavage of C3 by C3 convertases leads to the generation of a second major fragment, C3b, which acts as an opsonin facilitating the removal of bacteria and cell detritus by phagocytic cells [49] . Finally, the formation of C5b by cleavage of C5 initiates the assembly of a multimolecular complex, the MAC (C5b-9), that perforates membranes of bacteria and nucleated cells and causes rapid cell lysis and death [45, 50, 51] . Recently, a second initiation mechanism of the alternative activation pathway was described, termed the properdin pathway [52] . Properdin is capable of recognizing several DAMPs and PAMPs on foreign and apoptotic cells, thus allowing C3 convertase assembly on the target surface [32, 52] . Properdin also functions as a stabilizer for C3 convertase complexes of the alternative pathway. In addition to properdin, a fifth complement activation pathway has been described, which identified the clotting factor thrombin as a C5 convertase. This notion was supported by the observation that thrombin is capable of generating C5a in the absence of C3, thus providing a direct link between the complement and coagulation system [53, 54] . Traumatic brain injury (TBI) induces a profound inflammatory response that contributes to brain edema, neuronal cell death, and adverse outcome [55] [56] [57] . Posttraumatic activation of the complement cascade has been shown to play a pivotal role in the development of secondary brain injury (Table 1) [10, 12, 23, 24, 58, 59] . Multiple experimental Table 1 Insights from experimental complement inhibition based on genetically engineered mice and pharmacological approaches in models of traumatic brain injury (TBI). and clinical studies have revealed elevated levels of complement components and complement activation fragments in serum, cerebrospinal fluid (CSF), and brain parenchyma after head injury [12, 23, 60, 61] . Intracerebral complement deposition after TBI derives either from an altered permeability of a dysfunctional blood-brain barrier (BBB), or from posttraumatic biosynthesis of complement components by resident and infiltrating cells of the central nervous system (CNS) [12, [62] [63] [64] . Most studies have focused on the central complement component C3, and on the potential neuroprotective effects of inhibiting C3 convertases, the level at which the three main activation pathways merge, thus inhibiting downstream complement activation. Clinical studies revealed elevated C3 levels in the CSF of patients with severe TBI [65] . Experimental brain injury models described intracerebral PMNL infiltration and concomitant accumulation of complement C3 in cortical and hippocampal brain sections after experimental TBI in rats [66] . In those studies, C3 accumulation was significantly related to places of intracerebral cell death and to increased intracerebral myeloperoxidase activity [66] . In accordance with these findings, C3-deficient mice were found to have lower neutrophil extravasation and cerebral lesion volumes in a freeze model of brain injury [67] . In light of the central role of C3 and downstream complement activation fragments in the pathophysiology of TBI, much emphasis has been recently devoted to elucidating therapeutic aspects of C3 convertase inhibition, in various experimental model systems [68] [69] [70] [71] [72] . Genetically engineered mice, either deficient in the C3 gene, or with transgenic CNS-restricted overexpression of Crry -a soluble inhibitor of C3 convertases in mice-showed a significant extent of neuroprotection after brain injury, compared to wild-type animals [67, 70] . The GFAP-sCrry transgenic mice showed a significantly improved neurological outcome and an attenuated extent of posttraumatic BBB dysfunction in a model of closed head injury [70] . Based on these insights, the concept of Crry-mediated neuroprotection was extrapolated to a pharmacological approach, by posttraumatic injection of a recombinant chimeric Crry-Ig molecule in the same model of closed head injury [71] . The systemic injection of Crry-Ig during an early therapeutic "window of opportunity" within one hour to 24 hours after trauma resulted in a significant neurological improvement and reduced extent of neuronal cell death, compared to vehicle-injected control mice [71] . A similar therapeutic approach was tested in a fluid percussion model of brain injury, using recombinant Vaccinia virus complement control protein (VCP), a potent inhibitor of alternative and classical pathway C3 convertases [69, 72] . In these studies, the intracranial administration of VCP mediated neuroprotective effects related to posttraumatic preservation of spatial memory, as compared to vehicle-injected controls [69, 72] . Further therapeutic approaches were designed to more specifically target "key" effector components of complement activation, such as the anaphylatoxin C5a and its receptor (C5aR, CD88) [29, 67, 73, 74] . In addition, more attention was recently devoted to target specific pathways of complement activation exclusively, in order to overcome the potentially deleterious effects of a complete "shut-down" of complement activation at the central C3 level. This notion is based on the fact that complement also mediates neuroprotective effects in the injured brain, as e.g. shown by a dose-dependent protection of glutamate-induced excitotoxicity against neurons by the C3derived proteolytic fragment, anaphylatoxin C3a [75] , and by C3a-mediated induction of nerve growth factor (NGF) by microglia [76] . Based on the recent concept of a "dual role" for complement in the pathophysiology of brain injury, by promoting both early neurotoxic and late neuroreparative mechanisms after TBI [12, 77, 78] , the exclusive targeting of selected complement pathways was given more consideration, as opposed to the "pan" inhibition at the C3 convertase level [79] [80] [81] [82] . Among these, the targeted inhibition of the alternative pathway has drawn particular attention in recent years [79, 80, 83] . Factor B, the "key" component of the alternative pathway, was previously reported to be significantly elevated in the intrathecal compartment of patients with severe TBI [65] . Experimental studies on factor B-deficient mice (fB-/-), which are devoid of a functional alternative pathway, revealed significant neuroprotection after closed head injury, in conjunction with a decreased extent of posttraumatic complement activation [79] . These positive findings derived from studies in gene knockout mice were extrapolated into a pharmacological approach, using a neutralizing monoclonal anti-factor B antibody (mAb1379) in the same model system [80] . The postinjury injection of mAb1379 led to significantly attenuated extent of complement activation and anaphylatoxin C5a generation, and was associated with an improved neurological recovery and reduced neuronal cell death after experimental closed head injury [80] . These data imply an important role of the alternative complement pathway in contributing to the delayed neuropathology after TBI, and provide strategic opportunities for therapeutic targeting of alternative pathway molecules as a potential future pharmacological strategy. An additional avenue of research has been focusing on the terminal complement pathway, or "membrane attack" pathway, which results in cellular lysis by the MAC/C5b-9 [51, 84, 85] . In clinical studies, elevated levels of activated soluble MAC/C5b-9 were detected in the CSF of severely head-injured patients [62] . Moreover, the extent of intrathecal complement activation was associated with secondary cerebral insults in TBI patients, including post-injury BBB dysfunction [10, 62, 64] . Experimental studies have revealed that the intracerebroventricular injection of MAC induced a marked upregulation of adhesion molecule expression and leukocyte infiltration in the subarachnoid space and cerebral parenchyma [84] . In addition, MAC injection into hippocampus evoked seizures and neurocytoxic effects in rats [85] . Local MAC deposition in the injured brain was demonstrated in experimental models [86] and in injured human brains [87] . The complement regulatory molecule CD59 represents the main controlling molecule of MAC formation and an essential protector from neuronal cell injury after complement activation [51, 88] . Neurons express CD59 constitutively, as a protective mechanism from autologous "innocent bystander" cell lysis after complement activation in the brain [51, 89] . However, the posttraumatic activation of phosphatidyl-inositol-specific phospholipase C (PI-PLC) after traumatic brain injury renders neurons vulnerable to MAC-mediated lysis by shedding of the glycosyl-phosphatidyl-inositol (GPI)-anchored glycoprotein CD59 from neuronal membranes [88, 90] . A recent experimental study on closed head injury in mice lacking the gene for Cd59a (CD59a -/-) revealed increased susceptibility to brain injury in CD59a -/mice, compared to wild-type littermates [88] . In fact, head-injured CD59a -/mice showed increased neuronal cell death in tissue sections assessed by TUNEL histochemistry, in conjunction with elevated serum levels of neuron specific enolase (NSE), an indirect marker of neuronal injury [88] . These data corroborate the crucial role of the complement regulatory molecule CD59 in protecting neurons from complement-mediated lysis, and emphasize the impact of the terminal complement pathway in contributing to the pathophysiology of delayed neuronal cell death after TBI. Until present, there is a lack of specific pharmacological therapy designed to avoid induction of secondary brain injuries and delayed neuronal cell death [91] . There have been some significant advances in the field of therapeutic complement inhibitor development, in recent years [43, 74, [92] [93] [94] . While some of these inhibitors have been successfully tested in experimental head injury models (Table 1) [67, 68, 71, 80] , the "bench-tobedside" extrapolation to clinical applications in headinjured patients has yet to be accomplished [91] . Severe blunt chest trauma with associated pulmonary contusions is characterized by a robust inflammatory reaction which can result in exacerbated lung injury, acute respiratory distress syndrome (ARDS), multiple organ failure, and death [95] [96] [97] [98] [99] . Activation of alveolar macrophages and recruitment of neutrophils into the interstitial and alveolar compartments are followed by the release of an arsenal of proteinases and oxidants causing leakage of the pulmonary microvasculature and destruction of the alveolar epithelium [100] [101] [102] [103] . Various experimental models of lung injury could yield important insights into the critical role of complement activation products, particularly anaphylatoxin C5a, in the pathophysiology of trauma-induced lung inflammation and progressive alveolar injury [28, [104] [105] [106] . Elevated levels of C5a have been described in broncheoalveolar fluid samples from patients with acute lung injury [28, 107, 108] . When C5a was applied intratracheally in rats exposed to an IgG immune complex model, increased intrapulmonary generation of chemokines, accumulation of neutrophils and changes in vascular permeability could be detected [106] . The protective effects of anti-C5a were further corroborated by the observation that the antibody also suppressed release of tumor necrosis factor (TNF) into bronchoalveolar lavage [109] . Furthermore, C5a was shown to be required for TNF-dependent upregulation of intercellular adhesion molecule-1 (ICAM-1), an essential endothelial adhesion molecule required for neutrophil migration [109] . Czermak and colleagues demonstrated that both the in vitro and in vivo blockade of C5a led to significantly reduced production of CXC and CC chemokines [110, 111] . A proposed model for the current understanding of C5a-mediated inflammatory pathophysiology of acute lung injury is depicted in Figure 2 . Anaphylatoxin C5a has been shown to induce the early release of proinflammatory cytokines by alveolar macrophages, such as TNF and interleukin (IL)-1β [104] . Interaction of endothelial adhesion molecules (e.g. ICAM-1) with their corresponding receptors on neutrophils (e.g. CD11b/ CD18) leads to adhesion and transmigration of neutrophils into the alveoli [104] . Furthermore, release of TNF and IL-1β can also function in an autocrine way and activate alveolar macrophages to generate chemokines [112] . Among these, the different chemokines have been shown to further mediate neutrophil infiltration [113] . Activated neutrophils, alveolar macrophages and epithelial cells release reactive oxygen species and proteinases that cause diffuse alveolar and microvascular damage, thus exacerbating acute lung injury [111] . The interaction of C5a with its receptors, C5aR (CD88) and C5L2, is crucial for mediating the pulmonary inflammatory response. Bronchial and alveolar epithelial cells have been shown to express the C5aR [114, 115] . Mice lacking the C5aR gene showed a decreased extent of pulmonary inflammation, as characterized by attenuated myeloperoxidase production by neutrophils and decreased vascular leakage [116] . Furthermore, the use of a specific C5aR antagonist led to similar attenuation of inflammation signs in immune complex-induced lung injury, indicating the C5aR as a predominant effector of the C5a-mediated inflammation in the lung [117] . A recent study could point out that the cellular responses induced by C5a/C5aR interaction are potentiated by a tight connection between complement and Fcγ receptors [118] . Both C5aR and FcγR are known to be expressed on alveolar macrophages [111] . Shushakova et al. found that C5a causes induction of the activating FcγRIII and suppression of the inhibitory FcγRII during lung injury resulting in a pro-inflammatory reaction. Genetic ablation of C5aR expression in mutant mice completely abolished C5a/C5aR-induced regulation of FcγRs and led to decreased intrapulmonary generation of TNF and neutrophil accumulation [118] . Taken together, C5a seems to have a broader critical function through FcγR regulation, thus augmenting inflammation in the lung. In contrast to the C5aR, the effects of C5a are limited by C5L2 that is co-expressed with the C5aR on many cells including neutrophils [119] . Besides of C5a, C5L2 can also bind C5a desArg and potentially additional complement fragments [120] . Gerard et al. could demonstrate a greater influx of inflammatory cells and an enhanced release of IL-6 and TNF in C5L2-deficient mice in the model of immune complex-induced lung injury [121] . This observation proposes an anti-inflammatory role of C5L2 in the lung that seems to counteract C5a/C5aR-mediated inflammation. The complement-induced pulmonary response after chest trauma has been suggested to depend on a delicate balance between pro-and anti-inflammatory transcription factors [111] . Alveolar macrophage activation is characterized by increased nuclear translocation of nuclear factor(NF)-B and activator protein-1 (AP-1) representing an initial event in the genesis of the inflammatory cascade [112, 122] . In contrast to NF-B and AP-1, the transcription factor STAT3 has emerged as a negative regulator of the inflammatory response [28] . Interestingly, C5a has been shown to be responsible for STAT3 activation in lungs and alveolar macrophages after immune complex-induced lung injury whereas no complement-dependence could be found for activation of AP-1 [122, 123] . STAT3 has been hypothesized to act as a transcriptional mediator for the anti-inflammatory cytokine IL-10, and might contribute to a negative feedback system in acute lung injury [28, 111, 124] . In addition to the above described "classic" lung injury models, a recent study has paid more attention to the immune response after experimental blunt chest trauma induced by a blast wave [104] . Flierl and colleagues reported complement activation after trauma-induced bilateral lung contusion in rats with C5a-dependent perturbations in neutrophil functions. Treatment with anti-C5a antibody abolished functional deficits in neutrophils and reduced intrapulmonary levels of leukocytes and of cytokines [104] . Taken together, there is evidence from various animal models that support a predominant role of C5a in initiating a cascade of inflammatory events during acute lung injury. If lung trauma is severe, activation of the innate immune system can lead to a dysregulated inflammatory response resulting in ARDS [125] . Elevated levels of C3a and C5a were measured in plasma of patients with ARDS [126] . In addition, experimental complement inhibition led to attenuated pathology in an animal model of lung injury [126] [127] [128] . Thus, it is tempting to speculate that C5a might act as a potential target for immunomodulation after chest trauma [74] , to avoid the deleterious effects of posttraumatic inflammation, which lead to ARDS, multiorgan failure, and death [97, 129] . Experimental models of musculoskeletal trauma demonstrated that the early posttraumatic inflammatory response is often accompanied by robust generation of complement activation products [66, 104, 105] . However, up to now, the involvement of the complement cascade in bone and cartilage trauma has only been marginally investigated [130] . In recent years, increased attention has been devoted to the investigation of the role of complement in bone biology and fracture healing [131] . Mesenchymal stem cells as progenitor cells of osteoblasts were shown to express the complement receptors C3aR and C5aR, and the complement regulator molecules, CD55 and CD59 [132] [133] [134] . Moreover, osteoblastic differentiation as a key aspect of bone formation and remodeling induces upregulation of a number of complement-related genes, like C1q, C4, C3aR, properdin, C1-inhibitor (C1-INH) and complement factor H [135] . Pobanz and colleagues reported the expression of a functional C5aR by a human osteoblast-like cell line and detected increased osteoblast IL-6 production after stimulation of these cells with C5a [136] . Furthermore, vitamin D3 has been described to regulate C3 production by murine osteoblastic cells both in vitro and in vivo [137] [138] [139] . Complement C3 was postulated to exhibit a modulating influence on the differentiation of bone marrow cells into osteoclasts [139, 140] . Additional studies pointed out that complement appears to be involved in the transformation of chondral precursors to bone tissue during the enchondral ossification process, involving both the classical and alternative pathway complement activation [141, 142] . Consequently, complement components were hypothesized to be also involved in the inflammatory response after musculoskeletal trauma, and in mediating induction of fracture repair processes [131] . A recent study revealed that the C5aR is expressed in fracture callus by differentiated osteoblast, chondroblast-like cells, and osteoclasts [143] . Since fracture healing is known to be delayed in case of additional trauma-induced injuries, it furthermore remains to be examined if systemic complement generation might be the initiator of this delayed recovery after musculoskeletal trauma [144] . In addition to the role in fracture healing, the effect of complement activation on cartilage destruction after joint injuries has been discussed in recent years [130] . Gene expression analyses demonstrated that chondrocytes express a broad range of complement components and complement regulatory proteins [145] [146] [147] . The origin of complement components in the synovial fluid remains a topic of debate [130, 148] . Aside from chondrocyte-induced biosynthesis, it appears that multiple other non-cartilaginous sources contribute to complement release in the inflamed joint, including synovial cells and infiltrating leukocytes [130] . We recently hypothesized that chondrocytes may release pro-inflammatory cytokines, express neoantigens and undergo enhanced apoptosis after cartilage injury [130] . However, until present, the involvement of the complement system in posttraumatic joint inflammation and the development of posttraumatic osteoarthritis remains poorly understood, and requires further research. The pathophysiology of musculoskeletal trauma and of skeletal muscle ischemia/reperfusion is summarized in Figure 3 . The oxygen deficit in major trauma, in conjunction with subsequent reperfusion of ischemic tissues has been recognized as a trigger of an intense inflammatory response that may cause damage both locally in the affected muscle and also in remote organs primary not involved in the ischemic insult [149] [150] [151] [152] . Complement activation and consumption represents a critical event in the early phase of limb ischemia/reperfusion (I/R) injury resulting in the release of potent complement fragments like C3a and C5a [150, 153, 154] . It has been suggested that binding of preexisting natural IgM antibodies to neoantigen expressed by hypoxic cells after interruption of the blood flow is responsible for the activation of the classical complement pathway that importantly contributes to skeletal muscle I/R injury [155] [156] [157] . This hypothesis is strengthened by the fact that mice genetically deficient of mature B and T cells and natural antibodies (Rag1 -/mice) show significant reductions of tissue damage in a model of hindlimb ischemia and reperfusion [155, 158] . Furthermore, muscle edema and secondary neutrophil accumulation in the lung, as signs of reperfusion injury, were attenuated in C1q -/and C4 -/mice deficient in central components of the classical complement pathway [159, 160] . Aside from the classical pathway, recent data indicate important involvement of the classical and the lectin pathway in skeletal muscle I/R injury [159, 161] . A protective effect was attributed to the complement regulatory molecules decay-accelerating factor (DAF/CD55), C1-INH, and soluble complement receptor type 1 (sCR1) after skeletal muscle reperfusion injury [162] [163] [164] . Moreover, a pivotal role of C5a in causing lung damage after hindlimb I/R was shown in an experimental study in rats [165] . In accordance with this observation, multiple markers of local and remote organ injury were markedly reduced in C5-deficient mice, and in mice treated with a neutralizing C5aR antagonist [74, [166] [167] [168] . In summary, complement activation appears to play a significant role in contributing to post-injury inflammation in musculoskeletal trauma, including fractures, cartilage injury, and skeletal muscle I/R injury. Polytrauma is characterized as a syndrome of multiple injuries with defined severity which leads to a massive systemic immune activation and to secondary dysfunction and failure of remote, initially uninjured, organs [1, [5] [6] [7] . Clinical studies have demonstrated that complement activation occurs in plasma of patients after major trauma, as early as at the time of presentation in the emergency department [169] [170] [171] . The extent of complement-mediated inflammation was correlated with injury severity, tissue hypoperfusion, and posttraumatic [171, 172] . Serum levels of C3 and C3a were identified as markers of injury severity and outcome in multiply injured patients [173, 174] . Moreover, expression profiles of complement regulatory molecules and of the anaphylatoxin C5a receptor (C5aR/CD88) appeared to be significantly altered in leukocytes of multiply injured patients during the early phase of polytrauma, compared to blood samples from healthy volunteers [175] . The expression profiles of CD46 (membrane cofactor protein; MCP), CD59, and C5aR (CD88) on neutrophils correlated inversely with the severity of injury, an observation which was attributed to an intriguing trauma-induced "complementopathy" in multiply injured patients [175] . Sepsis represents a lethal complication of major trauma, characterized by an uncontrolled complement activation, as determined by significantly elevated plasma levels of C3a, C4a and C5a [176] [177] [178] . The anaphylatoxin C5a appears to represent the central molecule in the development of the overwhelming inflammatory response in sepsis, and has been coherently described as "too much of a good thing" [179] [180] [181] (Figure 4 ). Blockade of C5a was linked to improved survival in different experimental models of sepsis [182] [183] [184] [185] . Persistent elevation of C5a during progressive sepsis was related to a posttraumatic immunparalysis with "shutdown" of crucial neutrophil functions, including a loss of chemotactic and phagocytotic activity, impairment of the oxidative burst, and disturbances in intracellular signaling pathways [48, 186, 187] . Recent studies corroborated an important contribution of C5a in modulating apoptosis in different cell types during sepsis. While apoptosis rates in neutrophils were shown to be significantly attenuated during sepsis, lymphocytes, thymocytes and adrenal medullary cells exhibited increased C5a-dependent susceptibility to programmed cell death [188] [189] [190] [191] [192] . The latter phenomenon was hypothesized to be responsible for impaired adreno-medullary catecholamine release predisposing the development of septic shock [191] . Excessive C5a levels during sepsis were furthermore associated with reduced myocardial contractility and cardiac output, a phenomenon described as "cardiomyopathy of sepsis" [193] . In general, multiple organs seem to be put at increased risk for C5a-mediated damage induced by an abrupt upregulation of the C5aR in a variety of tissues (heart, lung, kidney, liver, thymus) in early phases of sepsis [194, 195] . A recent study implied that C5a-mediated signaling through the two C5a receptors (CD88 and C5L2) contributes to adverse outcome from sepsis [196, 197] . In experimental models of sepsis, the blockade of C5a and its receptors has been shown to protect end-organ function and to improve outcomes, thus providing a future new avenue for pharmacological treatment of this detrimental complication of major trauma [198] [199] [200] [201] . Future studies will have to be designed to validate this promising notion in a clinical setting. In recent years, multiple experimental and clinical studies have substantiated the notion of "key" role of complement activation after major trauma in contributing to the deleterious pathophysiological sequelae in the injured brain, lungs, and musculoskeletal system. Complement activation furthermore significantly contributes to the mechanisms of systemic post-injury complications, such as I/R injury, sepsis, and multiple organ failure. Therapeutic options aimed at attenuating the inflammatory complications of major trauma are currently unsatisfactory, and research strategies have largely failed in extrapolation from "bench to bedside". Experimental data from recent animal studies highlight the potential for complement inhibitors aimed at targeting central complement components and specific complement activation products, as promising future pharmacological agents in patients with major trauma. In this regard, site-targeted complement inhibition by new generation chimeric molecules which link pharmacological inhibitors to the local site of complement activation and tissue deposition may represent the future pharmacological "golden bullet". These chimeric molecules act locally at the site of injury and inflammation, and thus avoid the unwanted negative and adverse effects of a systemic complement blockade. Clearly, there is a tremendous need for well-designed experimental studies to shed some further light into our understanding of the complement-mediated pathology of major trauma, with the hope of designing and implementing new clinical treatment strategies for severely injured patients in the near future.

Predicting Biological Functions of Compounds Based on Chemical-Chemical Interactions

Given a compound, how can we effectively predict its biological function? It is a fundamentally important problem because the information thus obtained may benefit the understanding of many basic biological processes and provide useful clues for drug design. In this study, based on the information of chemical-chemical interactions, a novel method was developed that can be used to identify which of the following eleven metabolic pathway classes a query compound may be involved with: (1) Carbohydrate Metabolism, (2) Energy Metabolism, (3) Lipid Metabolism, (4) Nucleotide Metabolism, (5) Amino Acid Metabolism, (6) Metabolism of Other Amino Acids, (7) Glycan Biosynthesis and Metabolism, (8) Metabolism of Cofactors and Vitamins, (9) Metabolism of Terpenoids and Polyketides, (10) Biosynthesis of Other Secondary Metabolites, (11) Xenobiotics Biodegradation and Metabolism. It was observed that the overall success rate obtained by the method via the 5-fold cross-validation test on a benchmark dataset consisting of 3,137 compounds was 77.97%, which is much higher than 10.45%, the corresponding success rate obtained by the random guesses. Besides, to deal with the situation that some compounds may be involved with more than one metabolic pathway class, the method presented here is featured by the capacity able to provide a series of potential metabolic pathway classes ranked according to the descending order of their likelihood for each of the query compounds concerned. Furthermore, our method was also applied to predict 5,549 compounds whose metabolic pathway classes are unknown. Interestingly, the results thus obtained are quite consistent with the deductions from the reports by other investigators. It is anticipated that, with the continuous increase of the chemical-chemical interaction data, the current method will be further enhanced in its power and accuracy, so as to become a useful complementary vehicle in annotating uncharacterized compounds for their biological functions.

Metabolism refers to a collection of chemical reactions in vivo, which keep an unceasing supply of matter and energy for living organisms to maintain life (e.g., growth and reproduction) [1] . These energy-using and energy-releasing chemical reactions catalyzed by enzymes are organized into many metabolic pathways. Some compounds/small molecules play major roles in these pathways and are vital for many activities essential for life. For example, during the digestion, the energy rich molecules (i.e. carbohydrate) are broken apart to provide energy, which is then used by cells to build up complex molecules from simple molecules, such as utilizing amino acids to synthesize new proteins that the body needs. Identifying the biological functions of compounds is an effective way to study the mechanisms of many basic biological processes [2] . On the other hand, small molecules are the cause, and the cure, for many diseases. For example, diabetes mellitus is a metabolic disease caused by insufficient or inefficient insulin secretary response and elevated blood glucose level [3] . Compounds such as sulfonylureas [4] , acarbose [5] , biguanides, thiazolidinediones [5] , and sitagliptin [3] have been used as effective drugs for diabetic therapy. Therefore, it is essential to annotate the bioactivities of compounds, which will benefit drug design and disease treatment. Besides the conventional biochemical experiments, computational methods are alternative ways to annotate the biological functions of compounds. In recent years, various bioinformatics and structural bioinformatics [6] tools were developed to address this issue, such as Quantitative Structure Activity Relationship (QSAR) [7, 8] , pharmacophore modeling [9] , molecular docking [10] , and Monte Carlo simulated annealing approach [11, 12] . Different from these methods, Lu et al. [1] and Cai et al. [2] analyzed the biological functions of compounds by mapping them to the corresponding metabolic pathway classes, which are strongly associated with the biological functions of compounds. The functional group composition was used to represent the compounds, and the Nearest Neighbor Algorithm and AdaBoost learner [13] were used to construct the prediction models by Cai et al. [2] and Lu et al. [1] , respectively. Both the two prediction methods achieved quite promising results on their own datasets. However, none of their datasets contained the ''multi-function'' compounds that belong to two or more metabolic pathway classes. Since these authors were only focused on addressing the singlelabel classification problem, their methods could not be used to deal with the ''multi-function'' compounds. Actually, according to KEGG [14] , among all the compounds with functional annotations, the ''multi-function'' compounds occupy about 8%. Particularly, these multi-function compounds may play some unique role intriguing to both basic research and drug development and hence are worthy of our special attention. Recently, the systems biology methods based on protein-protein interactions have been widely applied for predicting protein attributes [15, 16, 17, 18, 19] . These algorithms suggest that interactive proteins are likely to share the common biological functions [16, 17, 18, 19] , also more likely tending to have the same biological function than non-interactive ones [20, 21] . Likewise, we can assume that the interactive compounds may tend to share the common biological functions. In this study, the chemical-chemical interactions were retrieved from STITCH [22] (Search tool for interactions of chemicals), where the interaction unit consists of two chemicals and their interaction weight. The interaction weight (confidence score) represents the probability that the interaction occurs between the two chemicals concerned. The interactive compounds can be classified into the following three categories: (I) ones that participate in the same reactions; (II) ones that share the similar structures or activities; (III) ones with the literature associations [22] . In a metabolism system, chemical reactions are organized into many metabolic pathways, thus the compounds involved in the same reactions are in the same metabolic pathways. Similar structures or activity means that they share the similar functions, and hence they are likely to be in the same metabolic pathways. The co-occurrence of two compounds in many literatures suggests some kinds of direct or indirect relationships, indicating they have the potential to be in the same metabolic pathways. Accordingly, it is rational to suppose that the interactive compounds tend to participate in the same metabolic pathways. In this study, we proposed a multi-target model based on chemical-chemical interactions for predicting the metabolic pathways where compounds participate in. Our method sorts the possible metabolic pathways that are associated with the query chemical, providing a more comprehensive view of the biological effects of the compound. According to a recent comprehensive review [23] , to establish a really useful statistical predictor for a biological system, we need to consider the following procedures: (1) construct or select a valid benchmark dataset to train and test the predictor; (2) formulate the statistical samples with an effective mathematical expression that can truly reflect their intrinsic correlation with the attribute to be predicted; (3) introduce or develop a powerful algorithm (or engine) to operate the prediction; (4) properly perform cross-validation tests to objectively evaluate the anticipated accuracy of the predictor. Below, let us describe how to deal with these steps. The compounds were retrieved from public available database KEGG [14] (Kyoto Encyclopedia of Genes and Genomes) compound [ftp://ftp.genome.jp/pub/kegg/release/archive/ kegg/42/ligand.tar.gz] (release 42.0). Subsequently, these compounds were mapped to the following 11 major metabolic pathway classes that are strongly associated with the biological functions of compounds (http://www.genome.jp/kegg/pathway. Table 1 under the title of Group-I). From the 4,366 compounds of Group-I, 3,137 compounds were retrieved that can interact with any of the others as annotated by STITCH database [22] (see Table 1 under the title of Group-II). Of the 4,366 compounds of Group-I, 4,027 are involved in only one metabolic pathway class, 246 in two metabolic pathway classes, 54 in three metabolic pathway classes, 24 in four metabolic pathway classes, 9 in five metabolic pathway classes, 4 in six metabolic pathway classes, 2 in seven metabolic pathway classes, and none in eight or more metabolic pathway classes. Of the 3,137 compounds of Group-II, 2,820 are involved in only one metabolic pathway class, 226 in two metabolic pathway classes, 53 in three metabolic pathway classes, 23 in four metabolic pathway classes, 9 in five metabolic pathway classes, 4 in six metabolic pathway classes, 2 in seven metabolic pathway classes, and none in eight or more metabolic pathway classes. Note that since one compound may occur in more than one pathway class, the sum of the compounds over the 11 pathway classes in Group-I turns out to be 4,860, which is greater than 4,366. Likewise, the sum of the compounds over the 11 pathway classes in Group-II is 3,606, which is greater than 3,137. This is quite similar to the case of proteins with multiple location sites, as elaborated in [24, 25] . The chemicals interactions were retrieved from STITCH [22] , a large database of known and predicted interactions of chemicals and proteins derived from experiments, literature, databases, and so on. As mentioned in Introduction, there are three types of associations between two compounds in STITCH: (I) cooccurrence in reactions, (II) similar structures or activities, and (III) literature associations. In the downloaded STITCH chemicals interactions file: chemical_chemical.links.detailed.v2.0.tsv from http://stitch.embl.de/cgi/show_download_page.pl, there are 337,482 pairs of interactive compounds belonging solely to type I, 73,598 pairs solely in type II, 2,152,508 pairs solely in type III, 384 pairs in both type I and II, 120,936 pairs in both type I and III, 10,372 pairs in both type II and III, and 1,990 pairs in the three types, in total of 2,697,270 interactions. Each of the interaction is quantified by the interaction confidence score, which represents the likelihood that the interaction occurs. In this study, the interactions with both interactive compounds occurring in the 4,366 compounds of Group-I were extracted. As a result, 3,137 compounds with 75,949 interactions were collected to constitute the benchmark dataset of the current study (see Table 1 under the title of Group-II). Besides the 4,366 compounds (cf. Table 1 under the title of Group-I) with known metabolic pathway classes, there are 11,661 compounds without known metabolic pathway classes in KEGG. Among these compounds, 5,549 compounds that have annotated interactions with the compounds of the 4,366 compounds in STITCH were collected. Such 5,549 compounds are to form an independent dataset, being used to test our prediction method in hopes to acquire useful information for further investigation. As mentioned in Introduction, the interactive compounds tend to participate in the same metabolic pathways. Accordingly, for a query compound, the higher interaction confidence score with its interactive compound, the more likely they are to participate in the same metabolic pathway. The more its interactive compounds involving in a certain metabolic pathway, the more likely it is to participate in such metabolic pathway. Based on these points, we should count not only the number of compounds interacting with the query compound, but also the corresponding interaction scores. Thus, the desired predictor can be formulated via the following procedures. Suppose the training dataset contains n compounds, which are denoted as fC 1 ,C 2 ,:::,C n g. The 11 metabolic pathway classes (cf. Table 1 ) are expressed as fP 1 ,P 2 ,:::,P 11 g, where P 1 represents the 1 st metabolic pathway class (''Carbohydrate Metabolism''), P 2 the 2 nd metabolic pathway class (''Energy Metabolism''), P 3 the 3 rd metabolic pathway class (''Lipid Metabolism''), and so forth. Thus, the descriptor of metabolic pathway classes to which the compound C i belongs to can be formulated as P(C i )~½p i,1 ,p i,2 ,:::,p i,j ,:::,p i,11 T (i~1,2,:::,n; j~1,2,:: where Given a query compound C q , its interaction with the compounds in the training dataset can be defined as W (C q )~½w q,1 ,w q,2 ,:::,w q,i ,:: where w q,i represents the interaction confidence score between C q and C i . T is the transpose operator, and w q,i~0 if no interaction exists between them. Here, we did not consider the selfinteraction, therefore w q,i~0 when q~i. Accordingly, the likelihood that the query compound C q is involved in the j-th metabolic pathway class can be formulated by the following score which is the sum of the interaction confidence scores of C q with its interactive compounds in the training dataset by counting both the number of interactive compounds and the interaction confidence scores. Obviously, the higher the score of Eq. 4, the more likely C q is to be involved in the j-th metabolic pathway C j . Thus, for a given query compound C q , we can use Eq. 4 to calculate its 11 scores, with each associated with one of the 11 metabolic pathway classes. The class to which the compound C q most likely belongs should be the one with the highest score. In other words, the query compound C q will be predicted to belong to the mth metabolic pathway class if m~arg max j S(C q [j)jj~1,2,:::,11 where m is the argument of j that maximize the value of S(C q [j). Since the problem in this study is of multi-label classification, we intend to provide flexible information by predicting some candidate metabolic pathway classes for the query compounds, rather than just the most likely metabolic pathway class. Therefore, instead of Eq. 5, let us consider the following equation containing 11 scores in a one-column vector: where D ; is a descending operator that sorts the 11 scores of Eq. 4 for S(C q [j) according to the descending order (S 1 §S 2 § Á Á Á §S j § Á Á Á §S 11 ). If there is a tie among these scores, a random order will be made among those with a tie. Consequently, the predicted metabolic pathway classes for the query compound can be derived according to the descending order of Eq. 6; i.e., if S 1~S (P k [6), S 2~S (P k [1), S 3~S (P k [10) , then it follows that the query compound C q is involved in the 6 th metabolic pathway class (''Metabolism of Other Amino Acids'') will be ranked as the highest in the likelihood, that C q in the 1 st metabolic pathway class (''Carbohydrate Metabolism'') as the 2 nd , and that C q in the 10 th metabolic pathway class (''Biosynthesis of Other Secondary Metabolites'') as the 3 rd . The corresponding results thus obtained are, respectively, called the 1 st -order, 2 nd -order, and 3 rd -order predicted metabolic pathway classes. And so forth. In statistical prediction, the following three cross-validation methods are often used to examine a predictor for its effectiveness in practical application: independent dataset test, subsampling (such as 5-fold, 7-fold, or 10-fold cross-validation) test, and jackknife test [26] . In this study, the 5-fold cross-validation was employed to examine the performance of our method. The concrete procedures were that the training dataset were divided into five groups by splitting each of its subsets into five approximately equal-sized subgroups. Each of these five groups was in turn used as a testing dataset and the rest used as training dataset, thereby generating five different success rates, with their average representing the success rate by the 5-fold cross-validation. For the j-th order prediction, the accuracy W j was calculated by where M j is the number of the compounds whose j-th order predicted metabolic pathway class is one of the true pathway classes that the compounds are involved with, and N is the total number of compounds in the dataset. Such 11-order accuracies were used to evaluate our prediction method. It is obvious according to the definition of Eq. 7 that, the higher the value of W j with a smaller value of j, or the lower the value of W j with a larger value of j, the better the prediction quality will be by our method. In the dataset, the average number of metabolic pathway class that each compound is involved in is calculated as where E i is the number of metabolic pathway classes that the compound C i is involved with. Hence, another measurement -the likelihood that the first k order predicted metabolic pathway classes cover all the true metabolic pathway classes that the compound is involved in -can be formulated as Usually, k is the smallest integer equal or greater than the average number of metabolic pathway classes (H). It is obvious from Eq. 9 that the larger the value of L k , the better the prediction quality will be by our method. Given a query compound, according to the information of its interactions with the 4,366 compounds in Group-I ( Table 1 ) whose metabolic pathway classes are known, the likelihood of its belonging to each of the 11 metabolic pathway classes can be easily calculated according to Eq. 4. And the scores thus obtained were sorted according to a descending order (Eq. 6) to yield the predicted metabolic pathway classes according to their different ranks or orders. In this study, our method was evaluated by the 5-fold crossvalidation on the benchmark dataset that contains 3,137 compounds in Group-II of Table 1 . The 11-order prediction accuracies are shown in Figure 1 . The first order (most likely) prediction accuracy is 77.97%, and the last order (least likely) prediction accuracy is 0.38%, which indicates a quite good performance of our method. The average number of metabolic pathway classes with which each compound is involved is 1.15 (cf. Eq. 8), meaning that the average success rate by a random guess would be 1.15/ 11 = 10.45%, which is much lower than that by our method. Accordingly, the parameter k in Eq. 9 was set to (1.15+1) = 2; i.e., we may select the results of the first two orders of the predicted metabolic pathway classes for the query compounds. As we can see from Figure 1 , the accuracies of both the 1 st and 2 nd order predictions are higher than that of the random guess. According to Eq. 9 the metabolic pathway classes predicted by the 1 st and 2 nd orders have actually covered more than 80% of all the true metabolic pathway classes, suggesting that, of the results predicted by the 11 orders, more attention should be paid to those by the first two orders. Listed in Table 2 are the accuracies by each of the 11 prediction orders for the 3,137 compounds about their involvement in the 11 metabolic pathway classes using the 5-fold crossvalidation test. The highest accuracy achieved by the 1 st -order prediction was 80.96% for the 1 st metabolic pathway class (''Carbohydrate Metabolism''). And the results obtained by the 1 st and 2 nd prediction orders have covered 89.00% of the true metabolic pathway classes. The second highest accuracy by the 1 storder prediction was 78.77% for the 11 th metabolic pathway class (Xenobiotics Biodegradation and Metabolism), while the results obtained by the 1 st and 2 nd prediction orders have covered 87.00% of the true metabolic pathway classes. Both the two 1 st -order accuracies are higher than the overall 1 st -order prediction accuracy of 77.97%, and each of their combinations with the 2 nd -order predictions is also higher than the overall likelihood of 80.00%. As for the metabolic pathway classes with less compounds, such as ''Glycan Biosynthesis and Metabolism'' class that contains only 68 compounds in Group-I and 43 in Group-II (cf . Table 1) , the predicted accuracies were relatively not as good as the others. It is anticipated that with more experimental data are available in future for the compounds in these classes, the corresponding prediction success rates will be improved. Overall speaking, the aforementioned results are quite encouraging, indicating that our approach may become a useful tool to deal with this kind of very complicated systems. As stated in the Method section, the interactive compounds derived from STITCH tend to participate in the same metabolic pathways. For example, Table 3 lists the interactions of dihydrouracil with other compounds. Among the 32 interactive compounds, most of them appear in ''metabolism of cofactors and vitamins'' or ''metabolism of other amino acids'' or ''nucleotide metabolism'' pathway class (cf. Table 1 ) just like dihydrouracil. Dihydrouracil and uracil participate in pyrimidine metabolism pathway (belong to ''nucleotide metabolism''), where 5,6-dihydrouracil and NADP+ are catalyzed by dihydropyrimidine dehydrogenase (DPD) to form uracil and NADPH+H+ [14, 27] . They are also co-mentioned in many PubMed Abstracts such as [28, 29, 30, 31, 32, 33, 34, 35, 36, 37] . Another two interactive compounds -dihydrouracil and dihydrothymine share a very similar structure, the only difference is that dihydrothymine has a methyl at the 5th position of the hexatomic ring while dihydrouracil has not [38] . According to the prediction criteria, when dihydrouracil was treated as a query compound, the first three order predicted metabolic pathways that it participates in are ''nucleotide metabolism'', ''metabolism of cofactors and vitamins'' and ''metabolism of other amino acids'', respectively, which are consistent with the true metabolic pathways that it is involved in. Predicted results for the compounds with unknown metabolic pathway Encouraged by the quite promising results obtained by the 5fold cross-validation test on the benchmark dataset of the 3,137 compounds, we applied the method to the 5,549 compounds whose metabolic pathways are unknown as mentioned in the Materials and Methods section. The predicted results thus obtained are given in Table S1 . As discussed above, we selected the metabolic pathway classes obtained by the 1 st and 2 nd order predictions for these compounds, in hoping that the information thus obtained may provide useful clues for further investigations. Actually, it is interesting to see that many of our predicted results have proved to be reasonable according to the reports from other investigators. For example, N-acetylgalactosamine 4-sulfate and its interactive compounds with pathway information are shown in Table 4 . N-acetylgalactosamine 4-sulfate can bind to sulfate, glucuronic acid, galactose, xylose, fucose, Na(+), glycerol, and phosphate to form complex to perform the biological function [39] . In PubMed Abstracts, N-acetylgalactosamine 4-sulfate is comentioned with sulfate [40] , glucuronic acid [41] , galactose [42] , 39-phospho.pho. [43] , sugar-1-phosph. [44] , UDP-GlcNAc [45] , indole-3-glyce. [46] , N-acetyl-D-glucosamine [47] , and GDPmannose [44] . Besides, N-acetylgalactosamine 4-sulfate and Nacetyl-D-glucosamine share a similar structure and the difference is that N-acetylgalactosamine 4-sulfate has a sulfate at the position 4 of the ring while N-acetyl-D-glucosamine has not [38] . From these evidences, N-acetylgalactosamine 4-sulfate is supposed to participate in the same metabolic pathways as its interactive compounds. It can be seen from Table 4 that most of the interactive compounds of N-acetylgalactosamine 4-sulfate belong to the 1 st and 2 nd metabolic pathway classes. By considering all the interactions and the interaction confidence scores, it was predicted that Carbohydrate Metabolism (the 1 st class) and Energy Metabolism (the 2 nd class) would be the possible metabolic pathway classes that N-acetylgalactosamine 4-sulfate belongs to. Actually, as a carbohydrate, N-acetylgalactosamine 4-sulfate reacts with Chondroitin 4-sulfate to form hydrogen oxide and G12336 (i.e. (GalNAc) 2 (GlcA) 1 (S) 2 ), one kind of glycan which can participate in Carbohydrate and Energy Metabolism. Therefore, N-acetylgalactosamine 4-sulfate may also participate in Carbohydrate and Energy Metabolism. Another example is that cyclopropylamine in Table 4 has 23 interactive compounds with known pathway information. Cyclopropylamine, cyanuric acid, ammonia, N-cyclopropylammelide, c0761, hydroxyl radicals are in the same pathway -N-cyclopropylmelamine degradation [48, 49] , where N-cyclopropylmelamine first reacts with hydrogen oxide to form N-cyclopropylammeline and ammonia, and then N-cyclopropylammeline also reacts with hydrogen oxide to form Ncyclopropylammelide and ammonia. After that, N-cyclopropylammelide reacts with hydrogen oxide to form cyanuric acid, cyclopropylamine and hydroxyl radicals. Finally, cyanuric acid is transformed into hydrogen oxide and ammonia through cyanurate degradation. Cyanuric acid, N-cyclopropylammelide, and c0761 are all in the 11 th pathway class. Therefore, cyclopropylamine may also belong to the 11 th pathway class (Xenobiotics Biodegradation and Metabolism). For other interactive compounds, they are comentioned with cyclopropylamine in PubMed Abstracts, such as polyethylene [50] , 1-aminocyclopropane-1-carboxylic acid [51] , cyclopropanecarboxylic acid [52] , 3-hydroxyphenylacetic acid [53] , and acetophenone [54] . In Table 4 , most of the interactive compounds of cyclopropylamine belong to the 11 th metabolic pathway classes. According to above analysis, cyclopropylamine is suggested to participate in the Xenobiotics Biodegradation Metabolism, which was the 1 st -order predicted class for cyclopropylamine by our method. Accordingly, it is quite reasonable to expect that our method may provide useful information for further investigating into biological functions of compounds from the viewpoint of system biology. As indicated by the above discussion and analysis, the results derived from the 1 st and 2 nd order predictions should be considered as the candidates for the metabolic pathway classes with which the query compound may be involved. In view of this, biochemical experiments should be conducted by mainly focusing on the targets predicted by the 1 st and 2 nd order predictions. The results obtained by the last five order predictions can be ignored due to their very low likelihood (,2%). Consequently, the current prediction method can provide useful clues for further validation by experiments and expedite the research progress by prioritizing the targets concerned. It is instructive to note that for the 4,366 compounds in Group-I of Table 1 , there are still 1,229 compounds that can not be processed by the current method due to lack of the interaction information with other compounds within the dataset. It is expected that the problem can be solved by collecting as much chemical-chemical interaction information as possible from STITCH, which is a large-scale and well-maintained resource in chemical biology, including the interactions information for over 2.5 million proteins and over 74,000 small molecules in 630 organisms. With the continuous increase of the interactions information, the performance of our method will be further improved. Based on the chemical-chemical interactions information, a multi-target model was proposed for identifying the metabolic pathway classes with which a query compound is involved. Since some compounds may be involved with more than one metabolic pathway class, our method is featured by the capacity able to provide a series of potential metabolic pathway classes for each of the query compounds investigated, instead of only one metabolic pathway class. It is anticipated that our method may become a useful tool in helping annotate the compound for their biological functions. Table S1 Each order predicted metabolic pathway class for the collected 5,549 compounds without known metabolic pathway classes. The predicted metabolic pathway class code corresponds to the code in Table 1 . Among the 11 predicted pathway classes, the first 2 order predicted metabolic pathway classes should be paid more attention to. (PDF)

Identification and Characterization of a Novel Non-Structural Protein of Bluetongue Virus

Bluetongue virus (BTV) is the causative agent of a major disease of livestock (bluetongue). For over two decades, it has been widely accepted that the 10 segments of the dsRNA genome of BTV encode for 7 structural and 3 non-structural proteins. The non-structural proteins (NS1, NS2, NS3/NS3a) play different key roles during the viral replication cycle. In this study we show that BTV expresses a fourth non-structural protein (that we designated NS4) encoded by an open reading frame in segment 9 overlapping the open reading frame encoding VP6. NS4 is 77–79 amino acid residues in length and highly conserved among several BTV serotypes/strains. NS4 was expressed early post-infection and localized in the nucleoli of BTV infected cells. By reverse genetics, we showed that NS4 is dispensable for BTV replication in vitro, both in mammalian and insect cells, and does not affect viral virulence in murine models of bluetongue infection. Interestingly, NS4 conferred a replication advantage to BTV-8, but not to BTV-1, in cells in an interferon (IFN)-induced antiviral state. However, the BTV-1 NS4 conferred a replication advantage both to a BTV-8 reassortant containing the entire segment 9 of BTV-1 and to a BTV-8 mutant with the NS4 identical to the homologous BTV-1 protein. Collectively, this study suggests that NS4 plays an important role in virus-host interaction and is one of the mechanisms played, at least by BTV-8, to counteract the antiviral response of the host. In addition, the distinct nucleolar localization of NS4, being expressed by a virus that replicates exclusively in the cytoplasm, offers new avenues to investigate the multiple roles played by the nucleolus in the biology of the cell.

Bluetongue is a major infectious disease of ruminants caused by an arbovirus (Bluetongue virus, BTV) transmitted by biting midges (Culicoides spp.) [1] [2] [3] . Historically, bluetongue has been endemic almost exclusively in temperate and tropical areas of the world where the climatic conditions favour both the spread of the susceptible insect vector population and the virus replication cycle within the vector [4] . However, in the last decade BTV has spread extensively in several geographical areas including Southern Europe and also, unexpectedly, in Northern Europe causing a serious burden to both animal health and the economy [5, 6] . From a molecular and structural virology perspective BTV is one of the best understood animal viruses. BTV is a member of the Orbivirus genus, within the Reoviridae family, and possesses a doublestranded RNA genome formed by 10 segments (Seg-1 to Seg-10) of approximately 19200 base pairs in total [1, 3] . Until now, the BTV genome has been shown to encode for 7 structural and 3 non-structural proteins. The BTV genome is packaged within a triple layered icosahedral protein capsid of approximately 90 nm in diameter [1, [7] [8] [9] [10] . The outer capsid of the virion is composed by 60 trimers of VP2 and 120 trimers of VP5 [11] and differences within this outer capsid define the 26 BTV serotypes which have been described so far [12, 13] . The outer capsid proteins, and VP2 in particular, stimulate virus neutralizing antibodies which in general protect only against the homologous serotype [14] . The internal core is formed by two layers, constituted by VP3 (subcore) and the immunodominant VP7 (intermediate layer) [7] . Three minor enzymatic proteins, VP1 (RNA dependent RNA polymerase), VP4 (capping enzyme and transmethylase) and VP6 (RNA dependent ATPase and helicase) are contained within the core that is transcriptionally active in infected cells [15] [16] [17] [18] [19] [20] [21] . The BTV genome encodes also 3 non-structural proteins: NS1, NS2 and NS3/NS3a. NS1 and NS2 are highly expressed viral proteins and their multimers are morphological features of BTVinfected cells. Multimers of the NS1 protein form tubules (approximately 50 nm in diameter and up to 1000 nm in length) that appear to be linked to cellular cytopathogenicity [22] , while NS2 is the major component of the viral inclusion bodies. NS2 plays a key role in viral replication and assembly as it has a high affinity for single stranded RNA and possesses phosphohydrolase activity [23] . NS3/NS3a are glycosylated proteins involved in BTV exit. There are two isoforms of NS3: NS3 and NS3a with the latter lacking the N-terminal 13 amino acid residues [24] [25] [26] . Therefore, the segmented genome of BTV has been thought to be monocistronic (i.e. ten genome segments encoding for 10 proteins) for almost three decades [27, 28] . Segment 9 however, contains the open reading frame (ORF) encoding VP6 but also a smaller coding sequence in the position +1 reading frame that is present in BTV and some related Orbiviruses such as African horse sickness virus and others [29] . Bioinformatic analysis predicts that the BTV ''ORFX'' encodes for a protein of 77-79 amino acid residues. This putative ORFX is subject to functional constraints at the amino acid level and its level of conservation is higher compared to that of the overlapping VP6. In addition, the ORFX putative AUG initiation codon has a strong Kozak context suggesting that this protein might be translated by leaky scanning [29] . Alternative reading frames are expressed in a variety of RNA viruses and they can play fundamental roles in viral replication and virus-host interaction. In this study, we identified a previously unknown non-structural protein and characterized its biological properties. All experimental procedures carried out in this study are included in protocol number 5182/2011 of the Istituto G. Caporale Initially, the open reading frame expressing ORFX (NS4) was amplified by PCR from BTV-10 (GenBank accession number D00509) and cloned into the pCI Mammalian Expression Vector (Promega) resulting into pCI-NS4. The BTV-8 NS4 was cloned into the peGFP-N1 vector (Clontech), resulting in plasmid pNS4-GFP. pNS4 7-77 -GFP, pNS4 13-77 -GFP and pNS4 19-77 -GFP are mutants derived from pNS4-GFP expressing NS4 truncated of the amino terminal 6, 12 and 18 amino acid residues, respectively. pNS4 7-77 -GFP, pNS4 13-77 -GFP and pNS4 19-77 -GFP maintain the methionine and valine residues in position 1 and 2 of NS4. Note that BTV-10 and BTV-1 NS4 are 100% identical at the amino acid level. While BTV-8 and BTV-1 NS4 differ for a single amino acid residue in position 6. The set of BTV-1 and BTV-8 plasmids necessary to rescue these viruses in vitro by reverse genetics were obtained following the method recently published by Boyce and colleagues [26] . Briefly, total RNA was extracted from infected cells using Trizol (Invitrogen) according to the manufacturer's instructions. Each BTV genome segment was amplified by RT-PCR using the AccuScript PfuUltra II RT-PCR Kit (Agilent) from either BTV-1 or BTV-8 dsRNA preparations and the resulting PCR products were gel-purified (Qiagen) and cloned into either pUC57 (Fermentas) or pCI. Each BTV segment was cloned downstream of a T7 promoter and upstream of a BsaI or SapI restriction site. All of the mutants described in this study were obtained using the QuikChange II Site-Directed Mutagenesis Kit (Stratagene), according to the manufacturer's instructions. All plasmids used in this study were completely sequenced before use. Sequences of PCR primers used in this study are available upon request. Antisera used in this study included polyclonal rabbit antisera raised against BTV VP7, NS1, NS2, NS3 and ORFX (NS4) expressed in bacteria as Glutathione S-transferase (GST)tagged recombinant proteins (Proteintech Group, Inc.). Antiserum against BTV-1 NS4 was raised against a recombinant GST fusion protein including the entire NS4 protein expressed in bacteria. Polyclonal rabbit antiserum against BTV VP6 was kindly provided by Polly Roy as previously described [32] . Antibodies against B23 and c-tubulin were obtained commercially (Sigma Aldrich). BTV-8 (IAH reference collection number NET2006/04) was originally isolated from a naturally infected sheep during the 2006 outbreak in Northern Europe [33] . The virus was passaged once in KC cells and once in BHK 21 cells. The reference strain of BTV-1 was originally isolated at the ARC -Onderstepoort Veterinary Institute (IAH reference collection number RSArrrr/01) and was adapted to cell culture by passaging it twice in embryonated eggs and 9 times in BHK 21 cells. Both viruses were kindly provided by Peter Mertens. Virus stocks were prepared by infecting BSR cells at a multiplicity of infection (MOI) of 0.01 and collecting the supernatant when obvious cytpopathic effect (CPE) was observed. The supernatants were clarified by centrifugation at 500 g for 5 min and the resulting virus suspensions aliquoted and stored at 4uC for short term usage and at 270uC for long term storage. Virus titres were determined by standard plaque assays using BSR or CPT-Tert cells [34] . Bluetongue is a major infectious disease of ruminants caused by bluetongue virus (BTV), an ''arbovirus'' transmitted from infected to susceptible hosts by biting midges. Historically, bluetongue has been endemic almost exclusively in temperate and tropical areas of the world. However, in the last decade BTV has spread extensively in several geographical areas causing a serious burden to both animal health and the economy. BTV possesses a double-stranded RNA segmented genome. For over two decades, it has been widely accepted that the 10 segments of BTV genome encode for 7 structural and 3 nonstructural proteins. In this study we discovered that BTV expresses a previously uncharacterized non-structural protein that we designated NS4. Although BTV replicates exclusively in the cytoplasm, we found NS4 to localize in the nucleoli of the infected cells. Our study shows that NS4 is not needed for viral replication both in mammalian and insect cells, and in mice. However, NS4 confers a replication advantage to BTV in cells in an antiviral state induced by interferon. In conclusion, we have elucidated a possible route by which BTV can counteract the defences of the host. Sequence analysis 65 full-length segment 9 sequences representing 24 BTV serotypes were obtained from GenBank. Amino acid conservation plots and secondary structure predictions were obtained using the CLC Genomics Workbench (CLC, Aarhus, Denmark) software and bioinformatics tools available online (the PSIPRED server [http:// bioinf.cs.ucl.ac.uk/psipred], and the Network Protein Sequence Analysis (nps@) server, [http://npsa-pbil.ibcp.fr/]). Recombinant BTVs were rescued by reverse genetics as previously described [26] . Briefly, plasmids containing the genomic segments of BTV-1 or BTV-8 or resulting mutants were linearized with the appropriate restriction enzymes and then purified by phenol-chloroform extraction. Digested plasmids were used as a template for in vitro transcription using the mMESSAGE mMA-CHINE T7 Ultra Kit (Ambion), according to the manufacturer's instructions. ssRNAs were purified sequentially by phenol/chloroform extraction and through Illustra Microspin G25 columns (GE Healthcare Life Sciences), following the manufacturer's protocol. Monolayers of 95% confluent BSR cells grown in 12 well plates were transfected twice with BTV RNAs using Lipofectamine 2000 (Invitrogen). Firstly, 0.5610 11 (BTV-1) or 1610 11 (BTV-8) molecules of each of the BTV segments encoding VP1, VP3, VP4, NS1, VP6 and NS2 were diluted in Opti-MEM I Reduced Serum Medium containing 0.5 U/mL of RNAsin plus (Promega) and then mixed with Lipofectamine 2000 diluted in Opti-MEM I Reduced Serum Medium. After 25 min of incubation at room temperature, the mixture was added to the cells. 16 to 18 h after the first transfection, the cells were transfected as before but with all 10 BTV segments. 3 to 4 h after the second transfection the cells were overlaid with 2 ml of minimal essential media containing 1.5% agarose type VII and 2% FBS, and monitored for development of plaques. Finally, individual BTV rescued clones were picked through the agarose overlay and used to infect fresh BSR cells in order to obtain a virus stock. Where necessary, BTV dsRNA was extracted from infected cells using Trizol (Invitrogen). The ssRNA fraction was precipitated using lithium chloride, and the harvested dsRNA fraction was precipitated using isopropanol in the presence of sodium acetate. Growth curves of BTV recombinant viruses used in this study were derived in cells infected at a MOI of 0.05 and testing for the presence of infectious virus in supernatants collected at 8, 24, 48, 72 and 96 h post-infection. Virus growth was also assessed in cells in the presence of 1000 antiviral units/ml (AVU/ml) of interferon Tau (IFNT) or universal type I interferon (UIFN). Recombinant ovine IFNT was kindly provided by Tom Spencer. IFNT was produced in Pichia pastoris and purified as described previously [35] . Universal type I Interferon (UIFN) was obtained from PBL InterferonSource. BFAE cells and CPT-Tert cells were treated with 1000 AVU of IFN 20 h prior infection with BTV recombinants at a MOI of 0.1 (BFAE and CPT-Tert), 0.01 or 0.001 (CPT-Tert). Two hours after infection, the medium was replaced and the cells maintained in the presence of either IFNT or UIFN at the original concentration. Cell supernatants were collected at 24, 48 and 72 h post-infection, centrifuged for 5 min at 500 g in order to pellet cell debris and virus infectivity was subsequently titrated by endpoint dilution analysis on BSR cells. Viral titers were calculated by the method of Reed & Muench and expressed as log 10 TCID 50 /ml [36] . Each experiment was performed two to three times, each time in duplicate, using different stocks for each virus. CPT-Tert cells were plated in 24-well plates and treated for 20 h with 1000 AVU/ml of IFNT or UIFN and then infected with either BTV-1 or BTV-8 at different MOIs (0.1, 0.01 and 0.001). The medium was replaced 2 h after infection and the cells maintained in the presence of either IFNT or UIFN at the original concentration. At 72 h post-infection, the cells were washed once with phosphate buffered saline (PBS; pH 7.4) and stained for 16 h using a 0.5% crystal violet/10% formaldehyde solution. We used Image-Pro Plus (MediaCybernetics), in order to quantify in each well the percentage of the monolayer that was disrupted after BTV replication. Results were expressed as the percentage of destroyed monolayer by calculating for each well the following formula: (number of pixels above background: total number of pixels times) X 100. BSR cells were transfected with 0.6-1.8 mg of either pCI-NS4, pNS4-GFP or derived deletion mutants, using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. For western blot analyses of intracellular proteins, cells were lysed by standard techniques as described previously [37] . For viral pellet analysis, cell supernatants were collected and viral particles concentrated 200 times by ultracentrifugation as previously described [38] . Protein expression was assessed by sodiumdodecyl-sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and western blotting using the various antisera as indicated above. Membranes were incubated with a horse radish peroxidaseconjugated secondary antibody (GE Healthcare Life Sciences) and developed by chemiluminescence using Amersham ECL Plus Western Blotting Detection Reagents (GE Healthcare Life Sciences). Experiments were performed using BSR, BFAE, CPT-Tert or C6/36 cells cultured in two-well glass chamber slides (Lab-Tek, Nalge Nunc International). Cells were either transfected with appropriate plasmids or infected with various BTV strains at a MOI between 0.01 and 1.5. Cells were washed with PBS and fixed with 5% formaldehyde for 15 minutes. The fixed cells were then processed as described previously [39] and incubated with the appropriate antisera. Secondary antibodies were conjugated with Alexa Fluor 488 (Invitrogen, Molecular Probes) or Alexa Fluor 594 (Invitrogen, Molecular Probes). Slides were mounted using VECTASHIELD Mounting Medium with DAPI (49,6-diamidino-2-phenylindole, Vector Laboratories). Slides were analysed and images collected using a Leica TCS SP2 confocal microscope. BSR cells were infected with BTV-1, BTV-8 or the corresponding deletion mutants at a MOI of 0.05 in 35 mm dishes. At 24 h post-infection, cells were fixed using cold 2.5% gluteraldehyde and 1% osmium tetroxide. Cells were subsequently pelleted through 1% SeaPlaque agarose (Flowgen), dehydrated using a graded alcohol series and embedded in Epon 812 resin, followed by cutting and analysis in a Joel 1200 EX II electron microscope. Animal experiments were carried out at the ''Istituto G. Caporale'' (Teramo, Italy) following local and national approved protocols regulating animal experimental use. Study 1. Litters of 3-day old NIH-Swiss mice (n = 8-12), were inoculated intra-cerebrally with 10 3 TCID 50 of either BTV-1, (2/2) 129/Sv], were inoculated intraperitoneally with 100 PFU of either BTV-1, BTV-8, BTV-1DNS4 or BTV-8DNS4. For each virus, two groups (n = 5) of mice were inoculated using two different virus preparations. Survival plots were constructed using data collected from two experimental groups (n = 10) with the exception of a mock infected group that was constituted by a single group of 6 mice. Formalin-fixed and paraffin-embedded brains tissue sections from inoculated (and mock inoculated) mice were used in immunohistochemistry. Sections (4-6 mm) were examined for the presence of BTV NS4 using a polyclonal NS4 antiserum and the EnVision (DAKO) detection system. The BTV genome is formed by 10 segments. Segment 9 contains an open reading frame (ORF) between nucleotides 182 and 418 ( Figure 1A ) in position +1 with respect to the major ORF expressing VP6 [29] . In silico analysis showed that this extra ORF is highly conserved and encodes a putative protein of 77-79 amino acid residues. A stretch of 11 basic amino acid residues is present in the N terminal portion of the protein (residues 3 to 20). In Figure 1A) . We generated a polyclonal antiserum towards ORFX, in order to assess whether BTV expressed this previously uncharacterized protein. We detected ORFX in BSR cells infected with either BTV-8 or BTV-1 by western blotting ( Figure 1B) . Controls included BSR cells transfected with plasmids expressing ORFX either in its native form, or with eGFP fused to its C terminus. These data confirm that BTV expresses a protein encoded by an alternative reading frame located in segment 9. We subsequently investigated whether ORFX was a structural or non-structural protein. We infected BFAE cells with BTV-1 and analysed supernatants (containing viral particles) and total cellular protein extracts. Unusually for mammalian cell lines, BFAE cells show very little BTV induced cytopathic effect (CPE), thus facilitating the efficient discrimination between all BTV proteins present in the cellular fraction and the structural proteins present in purified and concentrated viral particles released from infected cells. By western blotting, we detected NS1 and ORFX in the cellular fraction, while VP7 was abundantly present in the viral fraction (concentrated by ultracentrifugation) and barely visible in the cellular fraction ( Figure 1C) . We obtained the same results by infecting C6/36 mosquito cells (data not shown). We detected VP6 in both the cellular and the viral fraction ( Figure 1C) . Interestingly, unlike VP7, VP6 appeared to be relatively more abundant in cell lysates compared to the viral pellets, suggesting that there is an intracellular pool of this protein that is not incorporated in the BTV virions. The absence of ORFX in the viral pellet strongly suggested that this is a non-structural protein expressed by BTV. In light of these data, we designated this protein NS4. NS4 was also expressed in vivo, as shown by immunohistochemistry of brain sections of mice inoculated intracerebrally with BTV ( Figure 1D ). By confocal microscopy of cells transiently transfected with pCI-NS4, we observed that NS4 localized mainly in the nucleus (Figure 2A) where it showed a strong co-localization with the nucleolar marker B23 [40] . Importantly, cells infected with either BTV-1 or BTV-8 also showed a strong nuclear co-localisation between NS4 and B23 [40] (Figure 2B ). We also observed NS4 to localise in the nucleus of the C6/36 insect cells ( Figure 2C ). NS4 does not have a canonical nuclear localization signal (NLS) but possesses a stretch of basic amino acid residues, at the amino terminus portion of the protein, that could drive nuclear localization [41] (Figure 2D ). We constructed an NS4 expression plasmid (pNS4-GFP) and a series of deletion mutants (pNS4 7-77 -eGFP, pNS4 13-77 -eGFP and pNS4 19-77 -eGFP) lacking the 6, 12 and 18 amino terminal residues, respectively. pNS4-GFP and pNS4 7-77 -eGFP transfected cells showed a strong nuclear localization of NS4. On the other hand, NS4 showed a predominantly cytoplasmic localization in cells transfected with either pNS4 13-77 -eGFP or pNS4 19-77 -eGFP. These data suggest that the amino terminal basic domain of NS4 may play an important role in the nuclear localization of this protein. Interestingly, BFAE cells infected with BTV-1 revealed that NS4 expression was evident as early as 2 hours post infection, similar to that observed for other BTV structural and nonstructural proteins (Figure 3 ). The data above clearly show that a previously uncharacterized BTV protein, here referred to as NS4, is a non-structural protein that localises to the nucleolus of infected cells. Next, we generated by reverse genetics BTV NS4 deletion mutants in order to assess the requirement of this protein for viral replication. We generated a set of plasmids necessary for the rescue of BTV-1 and BTV-8 and engineered three mutations in the plasmids containing segment 9 of BTV-1 and BTV-8 such that the NS4 initiation codon was removed along with the introduction of two stop codons in the NS4 coding sequence. All the mutations introduced were designed in order to leave the VP6 amino acid sequence unaltered ( Figure 4A ). As a negative control for BTV rescue, we designed a VP6 deletion mutant with a premature stop codon incorporated into the VP6 coding sequence (position 79). As shown in Figure 4B , viable BTV1-DNS4 and BTV8-DNS4 were rescued with similar efficiency to the respective wild-type (wt) viruses, upon transfection of RNA transcribed in vitro from the appropriate plasmids representing the genomic segments of wt or mutated BTV-1 and BTV-8. As expected, BTV-1DVP6 and BTV8-DVP6 could not be rescued. We did not detect any variation in the migration pattern of dsRNA genomic segments extracted from all the wt or the NS4 deletion mutant viruses ( Figure 4C ). The RNA profiles of both the wt and DNS4 rescued viruses were identical to the corresponding profile of the stock viruses from which the segments were originally cloned. For each virus, segment 9 was completely sequenced in order to confirm the presence of the introduced mutations. We confirmed, by western blotting and confocal microscopy, that the DNS4 mutants do not express NS4 but express levels of VP7 and NS2 comparable to the parental wild type viruses. It was also evident that in BSR cells BTV-8 expresses lower amounts of NS4 relative to BTV-1 ( Figure 4D and not shown). However, BTV-1 replicates better than BTV-8 in these cells and differences in the steady-state levels of VP7 between these two viruses were also observed ( Figure 4D ). In cells infected by BTV1-DNS4 or BTV8-DNS4, we found by electron microscopy all the ultrastructural features of BTV-infected cells (e.g. viral inclusion bodies, NS1 tubules, viral particles) ( Figure 4E ). We next assessed the replication kinetics of the rescued viruses in a variety of mammalian and insect cell lines, including those corresponding to the natural hosts (sheep and cattle) and vector (midges). All subsequent experiments were performed using the rescued versions of the wt viruses as they represent a more homogenous population and are therefore more directly comparable to the rescued DNS4 viruses. Cells were infected with a MOI of 0.05 and supernatants were collected at various times postinfection ( Figure 5 ). No obvious difference was obtained in the replication of wt and DNS4 viruses, regardless of the cell lines used in the assay ( Figure 5) . Interestingly, the cell adapted BTV-1 viruses consistently grew more efficiently in vitro than the BTV-8 BTV-8 at a MOI of 0.05. At 48 h post-infection, cells were fixed and analyzed by immunofluorescence as for expression of NS4 as indicated in panel A. Scale bars correspond to 11 mm. (D) Confocal microscopy of CPT-Tert cells transfected with pNS4-GFP or the truncated mutants indicated above each panel. The red box corresponds to the first two amino terminal amino acid residues of NS4 that were maintained in all mutants. At 24 h posttransfection, cells were fixed and analyzed by immunofluorescence. Scale bars correspond to 18 mm. doi:10.1371/journal.ppat.1002477.g002 BTV NS4 confers a replication advantage to BTV-8, but not BTV-1, in mammalian cells treated with interferon BTV, like most RNA viruses, is a strong inducer of interferon, both in vivo in its natural hosts and in vitro [42] [43] [44] . Given that other RNA viruses express proteins that counteract the innate immunity of the host, we hypothesised that NS4 might aid BTV replication in the presence of interferon (IFN). We treated cells with two type I IFNs: IFN tau (IFNT) and universal IFN (UIFN). IFNT is secreted by the ruminant conceptus and it is intimately linked to pregnancy recognition signalling and possesses antiviral activity [45] while UIFN is an alpha interferon hybrid constructed from recombinant Human IFNs alpha A and alpha D, and is known to stimulate an antiviral response in a wide variety of mammalian cells. CPT-Tert cells were pre-treated with IFNT or UIFN for 20 h prior to infection with BTV-1 or BTV-8 (or mock infection) with MOIs ranging from 0.001 to 0.1. Both wt and the DNS4 mutants, destroyed 80 to 100% (depending on the MOI used) of the monolayer of infected cells in absence of IFN treatment ( Figure 6 ). On the other hand, pre-treatment with both types of IFN significantly reduced BTV-induced CPE. Interestingly, in the presence of IFN, BTV-8 wt consistently induced a more pronounced CPE than BTV8-DNS4. Conversely, only minor differences were observed in the CPE induced by both wt BTV-1 and BTV-1DNS4 in the presence of IFN ( Figure 6 ). Subsequently, we performed multi-step virus growth curves in order to further assess the replication of BTV wt and DNS4 in the presence or absence of IFN. CPT-Tert cells were treated with interferon, as described above, and infected at a MOI of 0.01 with wt and mutant viruses. At 24, 48 and 72 h post infection the cell supernatants were collected and the virus titrated in susceptible cells. BTV-8DNS4 consistently reached lower titres (approximately 10 to 25 fold) than wt BTV-8 in cells treated with 1000 AVU/ml of either IFNT or UIFN (Figure 7) . Similar to what was observed in the IFN protection assays, there was no discernable difference in the replication growth of BTV-1 and BTV-1DNS4 after treatment with either IFNT or UIFN. Similar patterns with both BTV-1 and BTV-8 wt and the DNS4 mutant viruses where observed when the input viruses were used at a MOI of 0.1 and 0.001 in CPT-Tert (data not shown), or in BFAE cells treated with UIFN and infected at a MOI of 0.1 (data not shown). We next ruled out that the mutations inserted in segment 9 of BTV-8DNS4 had a negative effect on VP6 expression (the other protein expressed by segment 9). As shown in Figure 8A , BTV-8 wt and BTV-8DNS4 express similar amounts of VP6, reinforcing the notion that the biological differences observed between these two viruses were indeed due to the expression of NS4. Therefore, the data presented so far suggested that either the BTV-1 NS4 was somewhat defective or that the influence of this protein on viral replication in the presence of IFN varies from strain to strain. In order to discern between these two possibilities, N in position 6) in order to render this protein identical to the homologous BTV-1 protein ( Figure 8B ). Both cytopathic protection assays and multistep growth assays clearly showed that BTV-8/1S9 replicated more efficiently than BTV-8/ 1S9DNS4 in the presence of IFN ( Figure 8C, D) . Similar results were obtained with BTV-8/1NS4, which replicated more efficiently than BTV-8DNS4 in cells pre-treated with IFN, while no major differences were observed between BTV-1/8S9 and BTV-1/8S9DNS4. Collectively, these data strongly indicate that the NS4 of BTV-1 is not defective and can function within the context of BTV-8. DNS4 BTV mutants are pathogenic in mice models of disease Next, we assessed the virulence of DNS4 BTV mutants in two murine models of bluetongue infection [46, 47] . 129sv IFNAR (2/2) mice, which are deficient in the type I IFN receptor, are susceptible to infection and disease induced by BTV inoculated by various routes [46, 48] . Newborn NIH-Swiss mice inoculated intracerebrally are also susceptible to BTV infection [47] . These models have been previously used to assess BTV virulence [47, 49] . In this study, we infected 129sv IFNAR (2/2) mice with either BTV-1, BTV-8 or the corresponding DNS4 mutants. No major differences were observed in the virulence of wild type and DNS4 viruses; all viruses employed in this study killed 100% of the inoculated mice by day 8 post-infection ( Figure 9 ). We also inoculated 3-day old NIH-Swiss mice intracerebrally with the same viruses as above. Once again, both wild type and DNS4 viruses were able to kill 100% of the inoculated mice with no major differences in the virulence observed ( Figure 9 ). In this study we have shown that BTV expresses a previously uncharacterised non-structural protein that favours viral replication in cells in an antiviral state. By constructing deletion mutants by reverse genetics, we showed that NS4 is dispensable for viral replication in vitro, both in mammalian and insect cells, and in vivo in murine experimental models. However, the coding sequence in the NS4 reading frame of segment 9 is highly conserved in BTV and in related Orbiviruses [29, 50] , suggesting that it must be essential for the maintenance of BTV in nature. Indeed, we have found that NS4 confers a replication advantage to BTV-8 in cells pre-treated with type I IFN. We found NS4 to have strong nucleolar localization, although it may shuttle between the nucleolus and cytoplasm and possibly carry out its biological functions in the latter. The nucleolus is a dynamic sub-nuclear structure that plays crucial roles in ribosome subunit biogenesis, the response to cellular stress and cell growth [51, 52] . Several examples of viral proteins targeting the nucleolus have been discovered in recent years [53] . The retroviral Rev and Rev-like proteins for example, shuttle between the nucleolus and cytoplasm, and function as post-transcriptional regulators of viral gene expression [54] [55] [56] [57] . One of the main functions of these proteins is to facilitate the export of unspliced viral mRNA (transcribed from the proviral DNA copy of the retroviral genome stably integrated in the cell genome) by simultaneously binding an RNA structure in the viral RNA and the karyopherin export factor Crm1 (chromosome region maintenance 1) [58] . Other RNA viruses (including those that replicate exclusively in the cytoplasm) have also been found to possess proteins that target the nucleoli. Examples include, among others, avian infectious bronchitis virus [59] , porcine reproductive and respiratory syndrome virus [60] , Newcastle disease virus [61] , Semliki forest virus [62] , dengue virus [63] , West Nile virus [64] , influenza virus [65] , avian reovirus [66] and encephalomyocarditis virus [67, 68] The reasons for the nucleolar targeting of many of these proteins have not always been entirely clear. 10 (TCID 50 /ml). In parallel, each virus preparation was also re-titrated by limiting dilution analysis to control that equal amounts of input virus was used in each experiment. This experiment was performed two times, each time in duplicate. doi:10.1371/journal.ppat.1002477.g008 The avian reovirus sA protein is a structural protein and is a major component of the inner capsid shell. Although the sA protein localises mainly in viral factories in the cytoplasm of infected cells, it also localizes in the nucleoli [66] . sA has a strong affinity for dsRNA and it may provide protection against the IFNinduced and dsRNA dependent PKR response. Interestingly, sA mutants that do not bind dsRNA are also unable to reach the nucleoli, suggesting that dsRNA binding and nucleolar targeting may be strictly linked [69] . BTV NS4 may also bind nucleic acids but, unlike the reovirus sA, we show strong evidence that NS4 is not a structural protein. Indeed, by western blotting we did not detect NS4 in viral particles but only in lysates of BTV infected cells. In addition, by confocal microscopy we did not detect NS4 in viral inclusion bodies but predominantly in the nucleoli of viral infected cells. We cannot exclude completely that small amounts of NS4, below the limits of detection of our western blotting analysis, are present in viral particles. The predicted structural features of NS4 resemble those of a transcription factor of the bZip family with a basic domain followed by a leucine zipper motif [70] . Thus, NS4 may function as a nucleic acid binding protein and either repress or enhance transcription of genes linked directly or indirectly to the IFN response of the cell. However, a BTV-8 recombinant virus (BTV-8DLZNS4) expressing an NS4 with all the 4 leucine residues forming the putative leucine zipper mutated (into either glutamine or serine) replicated as efficiently as BTV-8 wt in cells pre-treated with IFN (data not shown). Thus, more studies will be necessary to explore this possibility. The organization of VP6/NS4 ORFs in segment 9 of BTV mirrors that of NSP5/NSP6 in the rotavirus segment 11 [71] . The rotavirus NSP6 is not essential for virus replication but unlike the BTV NS4, does not localize in the nucleus of infected cells [72] . To date, limited information is available on the interplay between BTV and the host innate immune system. BTV has been recognized as a potent inducer of type I IFN in sheep [42] , cattle [43] and mice [73] . However, limited data have been available on how BTV induces the IFN response of the cell and, more importantly, what counteracting measures the virus utilises to overcome this response. Our data suggest that BTV may use NS4 to defend itself from the innate immune response of the host given that replication of BTV8-DNS4 in cells treated with IFN is 10 to 25 fold less efficient compared to wild type BTV-8. In addition, the cytopathic effect in cells treated with IFN is more pronounced when cells are infected by wild type BTV-8 compared to cells infected by BTV8-DNS4. Viruses have evolved a variety of strategies to evade the host innate immunity [74] . Other dsRNA viruses such as rotaviruses, use different mechanisms (which vary between strains and the type of infected cells) to modulate the type I IFN response. For example, rotaviruses use NSP1 protein to promote the proteasome-dependent degradation of IRF proteins [75] [76] [77] and mediate repression of NF-kB, resulting in a reduction of IFN induction [78] . Rotaviruses also induce shut off of cellular protein synthesis resulting from the detection of dsRNA by PKR which, in turn is responsible for phosphorylation and consequent inhibition of the eukaryotic translation initiation factor eIF2a [79] . The blocking of host cell protein synthesis is another likely strategy used by some RNA viruses to counteract the IFN response [67, 68] . BTV also blocks host cell protein synthesis early after infection, although the mechanisms underlying this phenomenon are not clear [27] . Interestingly, we found that BTV1-DNS4 replicated as efficiently as wild type BTV-1, even in cells treated with IFN. However, the NS4 of BTV-1 appears to possess the same biological properties of the NS4 of BTV-8. Indeed, a BTV-8 reassortant containing the entire segment 9 of BTV-1 (BTV-8/ 1S9) or a recombinant BTV-8 expressing an NS4 100% identical to the homologous BTV-1 protein (BTV-8/1NS4), maintained the phenotype of wt BTV-8. Thus, it is possible that the role played by NS4 in counteracting the IFN response of the host could vary between different virus strains. It is important to stress that the strain of BTV-8 that we used in this study has been passaged only a few times in culture (once in KC cells and three times in BHK 21 cells) after isolation from blood of an infected animal. On the other hand, BTV-1 was derived from the ''reference'' South African strain passaged twice in embryonated eggs and 9 times in BHK 21 . BTV-1 appears to grow slightly faster than BTV-8 in culture, especially at the early time points post infection. Thus, faster replication may help BTV1-DNS4 to escape the IFN response of the cell more efficiently, as already suggested for some strains of influenza, and this may render NS4 less critical in these in vitro assays [80] . More in vivo experiments will be needed in order to determine the role of NS4 in the interplay with the natural host of BTV infection. We observed no differences between wild type BTV-8 and BTV8-DNS4 in experimental mouse models, although it remains possible that differences could be identified in sheep. It is possible that NS4 is required for viral replication in insects, although we have established in this study that no differences are observed on the replication of the DNS4 mutants in insect cells in vitro. In conclusion, in the present study we have identified a previously uncharacterized non-structural protein of BTV. The identification of this highly conserved protein opens the way to understand finer details of virus-host interaction and pathogenesis. In addition, the distinct nucleolar localization, in a virus that replicates exclusively in the cytoplasm will offer new avenues to understand the various roles played by these organelles in the biology of the cell.

Interference of H-bonding and substituent effects in nitro- and hydroxy-substituted salicylaldehydes

Two intramolecular interactions, i.e., (1) hydrogen bond and (2) substituent effect, were analyzed and compared. For this purpose, the geometry of 4- and 5-X-substituted salicylaldehyde derivatives (X = NO(2), H or OH) was optimized by means of B3LYP/6-311 + G(d,p) and MP2/aug-cc-pVDZ methods. The results obtained allowed us to show that substituents (NO(2) or OH) in the para or meta position with respect to either OH or CHO in H-bonded systems interact more strongly than in the case of di-substituted species: 4- and 3-nitrophenol or 4- and 3-hydroxybenzaldehyde by ∼31%. The substituent effect due to the intramolecular charge transfer from the para-counter substituent (NO(2)) to the proton-donating group (OH) is ∼35% greater than for the interaction of para-OH with the proton-accepting group (CHO). The total energy of H-bonding for salicylaldehyde, and its derivatives, is composed of two contributions: ∼80% from the energy of H-bond formation and ∼20% from the energy associated with reorganization of the electron structure of the systems in question. [Figure: see text] ELECTRONIC SUPPLEMENTARY MATERIAL: The online version of this article (doi:10.1007/s00894-011-1044-1) contains supplementary material, which is available to authorized users.

and (3) of the 4-hydroxy-salicylaldehyde (in red, see Table 2 ); in parenthesis SESE for para and meta di-substituted benzene derivatives (in green, Tables 3 and 4 ). (2) and (3) of the 5-nitro-salicylaldehyde (in red, see Table 2 ); in parenthesis SESE for para and meta di-substituted benzene derivatives (in green, Tables 3 and 4 ). (2) and (3) of the 4-nitrosalicylaldehyde (in red, see Table 2 ); in parenthesis SESE for para and meta di-substituted benzene derivatives (in green, Tables 3 and 4 ).

Acute Respiratory Distress Syndrome Induced by a Swine 2009 H1N1 Variant in Mice

BACKGROUND: Acute respiratory distress syndrome (ARDS) induced by pandemic 2009 H1N1 influenza virus has been widely reported and was considered the main cause of death in critically ill patients with 2009 H1N1 infection. However, no animal model has been developed for ARDS caused by infection with 2009 H1N1 virus. Here, we present a mouse model of ARDS induced by 2009 H1N1 virus. METHODOLOGY PRINCIPAL FINDINGS: Mice were inoculated with A/swine/Shandong/731/2009 (SD/09), which was a 2009 H1N1 influenza variant with a G222D mutation in the hemagglutinin. Clinical symptoms were recorded every day. Lung injury was assessed by lung water content and histopathological observation. Arterial blood gas, leukocyte count in the bronchial alveolar lavage fluid and blood, virus titers, and cytokine levels in the lung were measured at various times post-inoculation. Mice infected with SD/09 virus showed typical ARDS symptoms characterized by 60% lethality on days 8–10 post-inoculation, highly edematous lungs, inflammatory cellular infiltration, alveolar and interstitial edema, lung hemorrhage, progressive and severe hypoxemia, and elevated levels of proinflammatory cytokines and chemokines. CONCLUSIONS/SIGNIFICANCE: These results suggested that we successfully established an ARDS mouse model induced by a virulent 2009 H1N1 variant without previous adaptation, which may be of benefit for evaluating the pathogenesis or therapy of human ARDS caused by 2009 H1N1 virus.

A novel influenza A (H1N1) virus of swine origin emerged among humans in Mexico during the spring of 2009 and rapidly spread worldwide [1] . The pandemic prompted the World Health Organization (WHO) to raise the alert level to the highest rating of six, the pandemic phase, within 2 months [2] . In August 2010, WHO officially declared that the disease was in the post-pandemic period [3] ; however, it is still circulating among humans, together with seasonal viruses. Although most influenza cases caused by 2009 H1N1 virus infection typically display mild upper respiratory tract syndrome, some cases progress to severe pneumonia and acute respiratory distress syndrome (ARDS) [4, 5] . Many studies have shown that ARDS caused by 2009 H1N1 virus results in 17.3-56% mortality [4, 6, 7, 8] , which was regarded as the major cause of death by 2009 H1N1 virus infection [9] . ARDS is the result of acute injury to lung tissue, commonly resulting from sepsis, trauma, and severe pulmonary infections [10] . Infectious factors, most of which are viruses, have become one of the most important causes of ARDS in humans [11, 12, 13] . Clinical cases and established animal models have revealed that the pathogenesis and pathological features of ARDS induced by different viral pathogens are distinct [14, 15] . However, knowledge of the pathogenesis of 2009 H1N1 virus, especially ARDS induced by 2009 H1N1 virus, is still limited and hinders therapeutic strategies. Therefore, it is necessary to evaluate the pathogenesis of ARDS caused by 2009 H1N1 virus infection in an appropriate animal model to assess potential therapies. Mice are a good model for evaluating the pathogenesis and antiviral therapy of influenza pneumonia, due to the general fidelity of the illness in mice to the human disease [16] . Moreover, a mouse model of ARDS caused by highly pathogenic H5N1 avian influenza virus infection has been well established [13] . The typical 2009 H1N1 virus, such as A/California/04/2009 (CA/04), can efficiently replicate in mouse lungs without prior host adaptation. However, it only causes moderate lung lesions and no mortality, even when inoculated at a high dose of 10 6 pfu [17, 18] . Thus, such typical 2009 H1N1 viruses may not be able to induce ARDS in a mouse model. In the present study, we used a virulent variant 2009 H1N1 virus, which was isolated from a pig and possessed a virulenceassociated HA-D222G mutation, to establish an ARDS mouse model. The model established here provides a useful tool to explore the mechanism of ARDS, as well as screening and therapeutic options. Six-week-old female mice were infected intranasally (i.n.) with 10 2.5 pfu SD/09 virus. Some of the infected mice showed signs of illness, such as altered gait, inactivity, ruffled fur, and anorexia on day 2 post-infection (p.i.). From day 2 p.i., the body weight of most mice significantly decreased ( Figure S1 ). By day 6 p.i., most mice presented with more severe clinical signs of respiratory disease, including labored respiration and respiratory distress, and most mice lost almost 20% of their initial body weight. On day 8 p.i., most mice were nearly unable to respond to exterior stimuli, and acute respiratory rates and labored respiration were observed (Video S1, and Video S2 for control). Approximately 60% of mice died between days 8 and 10 p.i. Gross observation of infected mice showed that the lungs were highly edematous, with profuse areas of hemorrhage and consolidation. No obvious gross lesions were observed in the kidneys, liver, spleen or brain of infected mice. Mice were infected i.n. with 10 2.5 pfu SD/09 virus, and three mice were euthanized on days 2, 4, 6, 8, 10 and 14 p.i., and the virus titers in viscera were determined. As shown in Figure 1A , the virus titer in the lung gradually increased between days 2 and 6 p.i., and reached a peak on day 6 p.i. The virus titers in the lung gradually decreased from day 6 p.i., and only one of three mice possessed detectable virus in the lungs on day 10 p.i. No viruses were detected in other organs, including heart, spleen, liver, kidneys, blood and brain, at the indicated time. These results indicate that SD/09 virus could replicate efficiently in mouse lung but did not cause systemic infection. As shown in Figure 1B , the effect of SD/09 viral infection on lung wet:dry weight ratio did not change significantly within 2 days p.i. However, a dramatic increase was observed from day 4 p.i., and reached a peak on day 8 p.i., which was nearly twice that observed in control group lungs (p,0.01). The change in lung wet weight:body weight ratio was similar to the change in lung wet:dry weight ratio ( Figure 1C ). The results indicated that the SD/09 virus could induce acute lung edema in mice. Kinetic observation of lung lesions of SD/09-virus-infected mice is shown in Figure 2 . On day 4 p.i., lung lesions were characterized by dropout of mucous epithelium and inflammatory cells adhering to the bronchiolar surface ( Figure 2C , D). On day 6 p.i., severe edema could be seen around blood vessels ( Figure 2E ); interstitial pneumonia was also observed that showed interstitial edema and thickening of the alveolar walls; and the alveolar lumen was flooded with detached alveolar cells, erythrocytes, and inflammatory cells ( Figure 2E , F). On day 8 p.i., the virus caused more severe interstitial pneumonia and peribronchiolitis, characterized by edema and extensive of lymphocytes, neutrophils and plasma cells around the area of bronchiolitis ( Figure 2G , H). Lesions in the lungs of infected mice were still severe on day 14 p.i., with extensive alveolar collapse, and remaining alveoli were filled with fibrin, desquamated alveolar cells, and inflammatory cells. Lymphocytes and alveolar macrophages were the predominant inflammatory cells observed at high magnification ( Figure 2J ). Masson's stain revealed that alveolar walls and spaces were filled with collagen fibers ( Figure 3A , B), indicating that proliferative fibroblastic lesions may develop. In comparison, lungs from mockinfected control mice had no apparent histological changes ( Mice were inoculated i.n. with 10 2.5 pfu SD/09 viruses; tissues were collected at indicated times p.i. and viruses were titrated in MDCK cells. Body weight, lung wet and dry weight were determined and recorded. The lung wet weight:body weight ratio and lung wet:dry weight ratio were calculated and used as an indicator of lung edema. *p,0.05, **p,0.01, ***p,0.001, comparison between ratios obtained from the virusinfected and control groups. Bars represent means 6 SD of data from three mice. doi:10.1371/journal.pone.0029347.g001 Immunohistochemistry revealed viral antigens in the epithelial cells of the bronchioles ( Figure 3D ), terminal bronchioles, and alveolar epithelial cells ( Figure 3E ). These data indicated that SD/ 09 virus could infect the epithelia of the lower airway and cause viral pneumonia in mice. As shown in Table 1 , virus-infected mice showed a slightly decreased partial pressure of arterial oxygen (Pa O2 ), saturation of arterial oxygen (Sa O2 ), and slightly increased partial pressure of arterial carbon dioxide (Pa CO2 ) from days 4 to 6 p.i. Most infected mice presented with severe clinical signs of respiratory distress on day 8 p.i., and blood gas analysis also showed that Pa O2 and Sa O2 dramatically decreased compared with the controls (p,0.05). These results suggested acute respiratory dysfunction and severe hypoxemia in virus-infected mice. The number of leukocytes in BALF from SD/09-infected mice showed an increase from day 4 p.i. ( Table 2 ). The BALF of virusinfected mice on day 6 p.i contained 1.1610 6 cells/ml and was significantly different from the 1.6610 5 cells/ml observed for PBSinoculated mice (p,0.001). These data indicate a dramatic increase in inflammatory cells in the lungs of SD/09-infected mice. To quantify the immune cell subpopulations responding to viral infection, we next determined cell differential counts in the infected lungs by Wright staining. Compared with PBS-inoculated animals, mice infected with SD/09 virus exhibited an increase of neutrophils from 4 days p.i., and the peak was 18-fold greater than that of the control group on day 8 p.i. Leukopenia was detected on day 4 p.i., and was statistically significant on day 6 p.i. (p,0.05); the lowest value appeared on day 8 p.i. (p,0.001). Furthermore, differential blood counts revealed that the number of lymphocytes sharply decreased in infected mice. The lowest number of lymphocytes observed occurred on day 8 p.i. (Figure 4B ), which dropped to ,20% of the control group number (p,0.001). To determine the cytokine responses that occur after SD/09 virus infection, we measured the levels of five cytokines and chemokines in lungs of infected mice on days 2, 4, 6, 8 and 10 p.i. As shown in Figure 5 , all five were significantly different between virus-infected and control mice. Interleukin (IL)-6 and IL-10 in the virus-infected mice reached peak levels as early as day 2 p.i. and were significantly higher than those of the control group (p,0.001). Interferon (IFN)-c, monocyte chemotactic protein (MCP)-1, and tumor necrosis factor (TNF) dramatically increased in mouse lungs on days 6-8 p.i. (p,0.001), consistent with the appearance of pulmonary lesions. These results showed that infection with SD/09 viruses resulted in elevated amounts of proinflammatory chemokines and cytokines in the lungs of mice. In the spring of 2009, a novel influenza A(H1N1) virus rapidly spread worldwide, resulting in the first influenza pandemic of the 21st century [1] . Critically ill cases caused by 2009 H1N1 virus retrospectively showed that most had progressed or died due to ARDS [4, 6] . However, the pathogenesis and therapeutic intervention of ARDS caused by 2009 H1N1 infection have still not been elucidated. Animal models of disease are important for characterizing pathogenesis and developing the preclinical evidence for revised approaches to ventilating patients with ARDS [19] . Here, we present a mouse model for the study of ARDS induced by SD/09 virus, a virulent 2009 H1N1 variant. Previous studies have indicated that typical 2009 H1N1 viruses such as CA/04 bind only to a-2,6-linked sialic acid (SA) receptor [17] , but only the a-2,3-linked SA receptor is found in the mouse respiratory tract [20] . Therefore, such a typical 2009 H1N1 virus may not be able to induce ARDS in mice. In fact, we used CA/04 virus to induce ARDS in mice; however, animals inoculated with a high dose of CA/04 virus (10 6.5 pfu) only showed moderate respiratory symptoms, and no lethality was observed (unpublished data). It has been shown that 2009 H1N1 virus possessing a D222G mutation in hemagglutinin (HA) could increase the pathogenicity in mice [21, 22] and binding to the a-2,3 SA receptor [23] . Moreover, clinical data indicate that such variants are only associated with severe H1N1 human infection [24] . Therefore, we suggest that the variant possessing the D222G mutation in HA can induce ARDS in a mouse model. The virus used in the present study was isolated from swine in 2009, and sequence analysis revealed that all the eight genes of the isolate had a close relationship with the 2009 H1N1 influenza virus circulating in humans. Notably, the swine isolate, SD/09, had a D222G mutation in HA. Compared with CA/04 virus (LD 50 .10 6 pfu), SD/09 showed significantly increased virulence in mice, with an LD 50 of 10 2.25 pfu, which was nearly identical to that of the mouse-adapted strain A/Hong Kong/415742Md/09 (LD 50 = 10 2.2 pfu) [17, 21] . Mice infected i.n. with 10 2.5 pfu SD/ 09 virus showed obvious respiratory symptoms, including visually prominent signs of respiratory distress and abdominal respiration, with approximately 60% mortality between days 8 and 10 p.i. The lungs of virus-infected mice were highly edematous, which was also demonstrated by dramatically increased lung wet:dry weight ratio. Pathological changes presented a progressive pattern, typically diffuse alveolar damage, interstitial and alveolar edema, neutrophil and macrophage-dominant inflammatory cellular infiltration, and areas of hemorrhage and necrotizing bronchiolitis. Arterial blood gas saturation is a key parameter of ARDS in humans [25] . In the present mouse model, Pao 2 and Sao 2 of infected mice were significantly lower than in the control group from day 2 p.i., especially on day 8 p.i., where these parameters sharply decreased, and most virus-infected mice began to die. These changes in arterial blood gas demonstrated that most infected mice developed severe hypoxemia consistent with of the appearance of clinical signs and lung lesions of ARDS. Previous studies showed that mice infected with typical 2009 H1N1 virus only exhibited mild interstitial inflammatory infiltration and limited alveolitis [18, 21] , whereas severe lung damage was found in the SD/09-infected mice, including severe edema around the blood vessels and bronchiolitis, and extensive inflammatory accumulation from 4 to 8 days p.i. At 10 days p.i., the surviving mice developed an irreversible fibrosis involving collagen deposition in alveolar walls and spaces, which was similar to that observed in human ARDS patients with 2009 H1N1 infection [26] . Our histopathological results were consistent with ARDS induced by other influenza viruses. Mice infected with mouse-adapted virus of the A/Puerto Rico/8/34 (H1N1), or high pathogenic H5N1 virus also showed a progressive series of pathological changes from interstitial pneumonia to diffuse alveolar damage [13, 27] . However, in contrast to highly pathogenic H5N1 virus, mice infected with SD/09 virus did not show viral spread to extrapulmonary organs. Immunohistochemical examination revealed the presence of viral antigens in the bronchioles, terminal bronchiolar epithelium, and alveolar epithelial cells. Perhaps SD/09 virus infection of the alveoli, particularly type II pneumocytes, rather than bronchioles, is a key to the development of ARDS. Type II pneumocytes are responsible for the production and secretion of surfactant to lower the surface tension of water and allow membrane separation, and insufficient pulmonary surfactant in the alveoli may result in alveolar collapse [28, 29] . The 1918 pandemic H1N1 and high pathogenic H5N1 viruses preferentially infect type II pneumocytes and alveolar macrophage in mice [15, 30] . Alveolar macrophages may play a critical role in disease pathogenesis, not through production of infectious virus but rather through the upregulation of proinflammatory cytokines that may further damage alveolar pneumocytes [31] . These phenomena suggest that viral cell tropism may determine the processes of ARDS. Pulmonary aberrant immune response is considered a significant feature of ARDS induced by 2009 H1N1 virus [19, 32] . In the present mouse model, the number of leukocytes observed in the BALF of virus-infected mice significantly increased compared with the control mice on day 8 p.i. Different counts in BALF showed that the proportion of neutrophils dramatically increased. These innate immune cells were capable of reducing the virus load in the lung [33] ; however, they could cause lung injury through direct or indirect mechanisms. Neutrophil oxidants and proteases can cause direct injury of cells in the alveolar-capillary membrane [34] . Neutrophils and macrophages can secrete copious amounts of chemokines and cytokines that can recruit more immune cells into lung tissues, and produce a ''cytokine storm'', one of the most important factors in the production of ARDS [35] . A retrospective cohort study of 74 2009 H1N1 patients found that higher levels of proinflammatory cytokines and chemokines in plasma were observed in the ARDS-death group compared with the survived-without-ARDS or the mild-disease groups [5] . Another study in critically ill patients with ARDS caused by 2009 H1N1 virus infection has shown that the hallmarks of disease severity were elevated levels of IL-6, IL-15, IL-8 and TNF-a [36] . We examined the levels of five cytokines and chemokines in infected mouse lungs and found significant differences between the virus-infected and mock groups. It has proved that high levels of IL-6 were able to mediate acute lung injury [37] , and had a negative correlation with the Pa O2 :Fi O2 ratio in severely affected patients with 2009 H1N1 virus infection [36] . Our data showed SD/09 viral infection induced high levels of IL-6 in mouse lung, which may also play an important role in the course of ARDS. Hagau etc. found the levels of TNF-a increased significantly in the 2009 H1N1-related ARDS patients [36] . In present study, TNF levels also dramatically increased in the lungs of virus-infected mice, and were consistent with the clinical symptoms and reached peak levels when mice began to die. In addition, high levels of IL-10, IFN-c and MCP-1 were also present in the virus-infected mouse lungs, similar to observations found in severely affected humans with 2009 H1N1 infection [38] . In summary, we successfully established an ARDS mouse model induced by a virulent 2009 H1N1 variant, which demonstrated key human ARDS clinical and pathological features, such as respiratory distress, low Pa O2 , exudative, proliferative and fibrotic lung, and high levels of inflammatory cells and cytokines. The mouse model may contribute to the study of the pathogenesis and therapy of ARDS induced by 2009 H1N1 virus. To determine LD 50 of SD/09 virus, eight 6-week-old female BALB/c mice per group were inoculated i.n. with 10 1 -10 6 pfu (50 ml) viruses and monitored for 14 days. The value of MLD50 was calculated using the Spearman-Karber method and expressed by pfu per MLD 50 [39] . We evaluated the pathogenicity of the virus in mice and found that it could efficiently replicate in the lungs of mice with high lethality (10 2.25 pfu per MLD 50 ). To determine the optimal dose of inoculation, 10 mice in each group were infected i.n. with 10 1.5 , 10 2.5 or 10 3.5 pfu viruses, and the signs, body weight, and mortality were monitored daily for each group for 14 days. Pilot experiments indicated that a dose of 10 2.5 pfu was optimal, because the course of the disease was prolonged and the mice presented with obvious signs of respiratory illness. BALB/c mice were lightly anesthetized and inoculated i.n. with 50 ml 10 2.5 pfu SD/09 virus in PBS. Mock-infected animals were inoculated i.n. with 50 ml PBS. At the indicated time, infected mice were sacrificed, and the parameters that present the course of the disease were determined. Twenty mice (10 infected with SD/09 virus and 10 inoculated with PBS) were used to investigate clinical signs and mortality for 14 days. Three mice were euthanized on days 2, 4, 6, 8, 10 and 14 p.i. and their organs were collected. The collected tissues were weighed, and 10% homogenates were prepared in cold PBS. The homogenates were centrifuged at 3000 rpm for 10 min to remove cell debris, and then the supernatants were 10-fold serially diluted for viral titer determination by plaque assay in MDCK cells. Virus titers were expressed as mean log pfu/g 6 standard deviation (SD). Three mice were euthanized on days 2, 4, 6, 8, 10 and 14 p.i., and the lungs were removed and weighed and then desiccated in an oven at 60uC for 72 h. The lung wet weight:body weight ratio and lung wet:dry weight ratio were calculated and used as an indicator of lung edema, as previously described [40] . Three mice were euthanized on days 4, 6, 8 and 14 p.i. The lungs were fixed in 10% buffered formalin, embedded in paraffin, sectioned, and stained with hematoxylin and eosin. Lungs on day 14 p.i. were also stained with Masson's trichrome. Lung tissue sections taken on day 6 p.i. were stained for influenza A virus antigens. An anti-influenza nucleoprotein monoclonal antibody (AA5H; Abcam, Hong Kong) was used to identify influenza A virus nucleoprotein in sections. Secondary antibody (Millipore, Billerica, MA, USA) against the primary antibody was labeled with horseradish peroxidase, and the color reaction was developed with a horseradish peroxidase reaction kit (diaminobenzidine-tetrahydrochloride; Sigma, St. Louis, MO, USA). Blood gas analysis was performed as previously described [13, 41] . Three mice were anesthetized with Zoletil (tiletamine-zolazepam; Virbac; 20 mg/g) on days 4, 6, 8, 10 and 14 p.i. Arterial blood samples were withdrawn into a heparinized syringe by percutaneous left ventricular sampling of lightly anesthetized mice that were spontaneously breathing room air. Blood gas analysis was immediately performed using a Vetstat Electrolyte and Blood Gas Analyzer (Idexx laboratories, Westbrook, MA, USA). Leukocyte counts in BALF were performed as previously described [42, 43] . Briefly, three mice were euthanized on days 4, 6, 8 and 10 p.i., and the lungs were lavaged twice in situ with the chest cavity opened by midline incision with a total volume of 1.0 ml saline (4uC) inserted through an endotracheal tube. The rate of recovery of BALF was not less than 90% for all animals tested. After the amount of fluid recovered was recorded, an aliquot of BALF was diluted 1:1 with 0.01% crystal violet and 2.7% acetic acid for leukocyte staining and erythrocyte hemolysis. The number of leukocytes in the BALF was counted with a hemacytometer under a microscope. For differential counts, the BALF samples from each mouse were stained with Wright stain, and the numbers of monocytes, neutrophil and lymphocytes were determined, on the basis of morphologic criteria, under a light microscope, with evaluation of at least 200 cells per slide. All slides were counted twice by different observers blinded to the status of the animal. Heparinized blood samples were collected on days 4, 6, 8 and 10 p.i. The total numbers of leukocytes and differential blood counts for three individual mice were analyzed using an automated hematology analyzer. IL-6, IL-10, TNF, IFN-c and MCP-1 levels were determined in lung homogenates using a cytometric bead array technique (BD Cytometric BEAD Array Mouse Inflammation Kit; BD Bioscience, San Diego, CA, USA) according to the manufacturer's instructions. Briefly, 50 ml mouse inflammation capture bead suspension and 50 ml PE detection reagent were added to an equal amount of sample standard dilution and incubated for 2 h at room temperature in the dark. Subsequently, samples were washed by adding 1 ml wash buffer and centrifugation at 2006 g at room temperature for 5 min. Supernatants were discarded and 300 ml wash buffer was added. Samples were analyzed on a BD FACSArray bioanalyzer (BD Bioscience) according to the manufacturer's instructions. Standard curves were prepared similar to the method above. Data were analyzed using BD CBA Software (BD Bioscience). Finally, the chemokine or cytokine levels were recorded as pg/ml homogenate. Data were analyzed by two-way analysis of variance using GraphPad Prism version 5.00 (GraphPad Software, San Diego, CA, USA). When a significant effect was observed, pairwise comparisons were performed using the Bonferroni post-hoc test. All data are reported as mean 6 SD.

True versus False Parasite Interactions: A Robust Method to Take Risk Factors into Account and Its Application to Feline Viruses

BACKGROUND: Multiple infections are common in natural host populations and interspecific parasite interactions are therefore likely within a host individual. As they may seriously impact the circulation of certain parasites and the emergence and management of infectious diseases, their study is essential. In the field, detecting parasite interactions is rendered difficult by the fact that a large number of co-infected individuals may also be observed when two parasites share common risk factors. To correct for these “false interactions”, methods accounting for parasite risk factors must be used. METHODOLOGY/PRINCIPAL FINDINGS: In the present paper we propose such a method for presence-absence data (i.e., serology). Our method enables the calculation of the expected frequencies of single and double infected individuals under the independence hypothesis, before comparing them to the observed ones using the chi-square statistic. The method is termed “the corrected chi-square.” Its robustness was compared to a pre-existing method based on logistic regression and the corrected chi-square proved to be much more robust for small sample sizes. Since the logistic regression approach is easier to implement, we propose as a rule of thumb to use the latter when the ratio between the sample size and the number of parameters is above ten. Applied to serological data for four viruses infecting cats, the approach revealed pairwise interactions between the Feline Herpesvirus, Parvovirus and Calicivirus, whereas the infection by FIV, the feline equivalent of HIV, did not modify the risk of infection by any of these viruses. CONCLUSIONS/SIGNIFICANCE: This work therefore points out possible interactions that can be further investigated in experimental conditions and, by providing a user-friendly R program and a tutorial example, offers new opportunities for animal and human epidemiologists to detect interactions of interest in the field, a crucial step in the challenge of multiple infections.

Numerous parasites species circulate simultaneously in natural populations. Many of them are able to infect a same host species and a host individual can therefore be infected by several parasites at the same time. These multiple infections are not only common in nature but usually more frequently encountered than infections by a single parasite [1] . Within a host individual, parasites can thus interact, either in a synergistic manner (parasite A favours infection by parasite B or worsens the symptoms caused by B) or in an antagonistic manner (parasite A decreases the infection risk by parasite B or reduces the symptoms caused by B) [2] . As these interactions can have important epidemiological, biological and clinical consequences (e.g., [3] [4] [5] [6] [7] ), detecting, understanding and evaluating them is essential to understand the phenomena and to control and manage infectious diseases. In recent years, the question of polyparasitism has attracted considerable attention [4, 8, 9] , although in reality the subject has a long history of experimental investigation under laboratory conditions [10, 11] . Many epidemiological studies have also been conducted on the main human pathogens, motivation for the study of polyparasitism being in particular driven by the urgency to understand the epidemiological and clinical consequences of infection by parasites potentially interacting with HIV and other emerging diseases [12] and the mechanisms of their interactions. A large amount of work indeed revealed interactions between HIV and tuberculosis, malaria, sexually transmitted diseases, and helminths (e.g., [6, [13] [14] [15] [16] [17] [18] [19] [20] ); as well as interactions between plasmodia parasites and helminths (e.g., [21] [22] [23] [24] ). Studies on animal hosts also revealed interactions between their parasites, with many studies on helminth communities (mammals: [4, 25, 26] , birds: [27] , fish: [28] ), and fewer on protozoan species (e.g., [29] ) or viruses (e.g., [30] ). Many diseases have been revealed to be affected by the presence of other disease-causing agents, altering the rates of species co-occurrence, levels of infection and disease severity. Parasite interactions have also been shown to affect the success of parasite vaccination strategies [31] and could be involved in disease (re)emergence [32] , reinforcing the interest of these studies. If laboratory experiments have clearly demonstrated that interspecific parasite interactions occur, often mediated by host immune responses [1, [33] [34] [35] , attempts to detect such effects in natural populations have generally been less successful. Indeed, detecting their existence on the field is not easy, due to complex networks of indirect effects making it difficult to infer underlying processes. Field studies are however essential as experimental systems are oversimplified and require an existing suspicion of interaction between the studied parasites. In addition, only studies in natural populations can give access to infection and co-infection probabilities. In other words, before studying their mechanisms in the lab, interactions of interest must be identified in the field. Main difficulties encountered in field studies are methodological. Many confounding factors can create statistical associations between parasites even if there is no true biological interaction between them, which may alter conclusions about the importance of interspecific interactions [36] [37] [38] [39] [40] . A similar transmission mode, for example, can alone increase the risk of co-infection. The excess of positive associations found in strongylid communities in domestic horses, ruminants and macropod marsupials is in particular likely to be due to the common habit of these hosts feeding on pastures contaminated with the larvae of a number of nematode species [41] [42] [43] . In addition, environmental, behavioural or host-specific factors can be associated with both types of infection and influence epidemiological and geographic patterns of infection and disease. Among such common risk factors, some have long been recognised, such as sexual behaviours for sexually transmitted diseases (e.g., [44] ), socio-economic status for infections particularly prevalent in poor regions such as helminth infection and malaria [45] , or age for many diseases (e.g., [46] ). As apparent associations between two infections may be due to common risk factors, they are crucial to identify and to take into account in the analysis. However, such confounding factors are difficult to control and few methods enable to take them into account. A variety of analytical approaches have been suggested to detect associations in parasite communities, primarily focusing on macroparasite (parasitic helminth) communities (e.g., [4, 39, 47, 48] ). However, they implicitly assume that the direction and strength of an observed association between parasite species reflects an underlying biological interaction, and their reliability to detect interactions has been recently questioned [49] . The adoption of a generalized linear mixed modelling (GLMM)-based approach has been rather suggested by Fenton et al. [49] (see also [50] ). Apparently more robust to detect interactions between macroparasites, this method has the advantage of offering the opportunity of taking into account the variance caused by other factors. Nevertheless, field data, particularly relating to microparasites, are most of the time serological (i.e. presence-absence data). Indeed, viral excretion is usually too short to make antigen detection an efficient tool to follow microparasites in natural populations, as host capture and sampling would have to be done exactly during the excretion period, especially during nonepidemic phases. Most field data are thus limited to observed frequencies of seronegative, seropositive and doubly seropositive individuals. In this context, the search for potential interactions between pairs of microparasites is traditionally done by calculating odds ratios in stratified data or by a Pearson's chi-square test of independence (e.g., [29, 51, 52] ). The latter compares the observed frequencies to the frequencies expected if parasites are independent, under the null hypothesis that the joint distribution of the cell counts in a 2-dimensional contingency table is the product of the row and column marginals. However, such methods ignore confounding factors and/or the possible simultaneous action or interaction of several of them. Significant associations detected in this manner can therefore be either true biological interactions or statistical associations, with no means of distinguishing the two. Alternative methods have been therefore proposed to determine the expected frequencies in a modified chi-square analysis. Some are based on the estimation of ''pre-interactive'' species prevalences [53] , which requires previous knowledge of dominance relationships between parasites species. Some others are based on log-linear models [e.g., [54] [55] [56] . In addition, another way to take risk factors into account is to include them in a logistic regression analysis and to determine whether parasite B status is still a predictor of parasite A status [52, 57] . However, the main drawback of methods based on log-linear or logistic regression models is that they are based on an asymptotic approximation of the deviance, which might not be relevant for small sample size data. In the present paper, we propose another method (termed the ''corrected chi-square'') to detect microparasite interactions from serological data, based on an adaptation of the Pearson's chisquare test. By combining logistic regressions and chi-square tests, we are able to calculate the expected frequencies of co-infected individuals if parasites are independent considering their risk factors, and to compare them to the observed ones. In a first step, we perform a theoretical comparison of the robustness of the corrected chi-square and the logistic regression approaches. In a second step, both approaches are applied to serological data obtained in natural populations of domestic cats to search for potential interactions between four feline viruses. The domestic cat is indeed an appropriate model to investigate such questions as its main viruses are well known and rather easy to survey on the field, and its natural populations, although very flexible in their social and spatial organisation, have been extensively studied [32, [58] [59] [60] [61] [62] . 1.1. Logistic regression analysis. A first way to test the interaction between two pathogens is to test the effect of the serological status to one virus on the probability of being seropositive to the other. A logistic regression was used for that purpose. The approach allows correcting for common risk factors by adding known or suspected risk factors as correction variables. The logistic regression model reads: Where F k denotes the k-th risk factor, p 1 is the probability of seropositivity to pathogen 1 and S 2 the serological status to pathogen 2. The coefficients a k (k = 0…K) and b are the coefficients of the logistic regression. The interaction between the two pathogens was tested using a likelihood-ratio test (LRT) testing H0: b = 0 vs H1: b=0. The asymptotic chi-square approximation was used to derive the Pvalue of the test of independence between the two viruses [63] . 1.2. Corrected Pearson's chi-square tests. The corrected chi-square approach is based on the idea that the coefficients of the logistic regression of the two viruses can be used to estimate the number of seronegative, single-and double-seropositive individuals expected if the two pathogens are independent. As the classical chi-square, the corrected chi-square compares the observed (O i,j ) and theoretical (E i,j ) numbers of individuals with different combinations of status (seropositive or seronegative) for the two pathogens using the chi-square statistic: where i is the status to pathogen 1 (0 for seronegative and 1 for seropositive) and j is the status to pathogen 2. To calculate the E i,j , for each pathogen taken separately, a logistic regression including K risk factors (see previous section) is run to estimate the probability of being seropositive for each individual (termedp p p,x for individual x and pathogen p, p[ 1,2 f g): Whereâ a p,k denotes the estimation of the regression coefficients for pathogen p and F k,x the value of the k-th risk factor in individual x. The theoretical contingency table is then deduced from these probabilities: For each pair of viruses, the distribution of the corrected chisquare was determined by a parametric bootstrap run as follows: Step 1: Estimated seropositivity probabilities (p p p,x ) are used to generate in silico serological data for both pathogens independently. Step 2: The corrected chi-square is calculated for this in silico dataset. Steps 1 and 2 were repeated 1000 times, leading to 1000 independent realisations of the corrected chi-square statistic under the null hypothesis of independence between the two pathogens. Two ways of calculating the P-value were derived from this procedure. P-value1 was estimated assuming that the corrected chi-square is proportional to a chi-square with one degree of freedom, the coefficient of over-(or under-) dispersion (ĉ) being defined by the mean of the bootstrapped corrected chi-square. P-value2 was given by the proportion of bootstrapped corrected chisquares which were smaller than the observed value. In principle, P-value2 is better (no assumption on the distribution of the Likelihood Ratio Test, LRT, is made), but requires running enough simulations, which may be long in some cases. P-value1 allows working with smaller numbers of simulations when simulation times are too long. The R program is available as supplementary file (File S2) and can be applied to any presence-absence data to calculate the corrected chi-square and the associated P-values. A tutorial example (File S3) illustrates its use step-by-step using an example dataset (File S4). The main criticism that could be made to the logistic regression approach is that it is based on the asymptotic distribution of the LRT. In practice, the chi-square approximation is true only for large datasets. In the present paper we investigated the robustness of the logistic regression to different sample sizes and numbers of correction risk factors. We also aimed to compare how robustness is affected by the type of risk factors considered (qualitative or quantitative). The same investigations were performed with the corrected chi-square test to compare the robustness of the two approaches. For that purpose, random seroprevalence datasets were generated, assuming independent viruses. Random data were always generated assuming that all individuals had an independent 0.5 probability of being seropositive for each pathogen. N F randomly generated risk factors were considered in the logistic regression for the two pathogens. By construction these factors have no effect (they are chosen independently of the serological status of the individuals) but from a theoretical point of view it is interesting to measure how their inclusion in the model can introduce biases depending on the approach. Randomly generated factors could be either qualitative or quantitative. For simplicity, qualitative factors had only two modalities, individuals having a 0.5 probability of being in each one. Quantitative factors were chosen for each individual randomly according to a standard normal distribution. To investigate how the nature of risk factors affects robustness, three scenarios were tested: i) all factors are qualitative; ii) all factors are quantitative and iii) half of the factors are quantitative while the other half are qualitative factors (mixed scenario). Our objective now was to understand how data characteristics (the number of individuals, n, the number of factors, N F and their type, scenario i, ii or iii) would affect the probability of wrongly concluding that there is an interaction between the two pathogens (type I error). For a given combination of these characteristics, a thousand random seroprevalence datasets were generated and we estimated the type I error associated to each approach as the proportion of random datasets for which the P-value was below 5%. 3. Application to cat data 3.1. Ethics Statement. The field work has been made by qualified people according to the French legislation. Accreditation has been granted to the UMR-CNRS 5558 (accreditation number 692660703) for the program. 3.2. The feline viruses. The Feline Immunodeficiency Virus (FIV), is a major non-traumatic cause of death in adult cats, and is associated with immunosuppression causing secondary infections [64] . This retrovirus can infect other felids, most of which are threatened or endangered species e.g., the European wildcat (F. s. silvestris) [64] [65] [66] . It is mainly transmitted by bites, through a direct horizontal mode [67] , principally during aggressive or sexual contacts [64, 68] . The Feline Herpesvirus (FHV) and the Feline Calicivirus (FCV) are responsible of upper respiratory tract disease, of concern in veterinary medicine [69, 70] . Both viruses are transmitted through 'amicable' contacts, by oral, nasal and ocular secretions during close interactions [71, 72] . FHV infected cats become asymptomatic carriers, but the latent infection can be reactivated by a stress (i.e., change of habitat, lactation or fights between males; [73] ). The Feline Parvovirus (FPV) infects all felids, as well as other carnivores [74] , and FPV infection may be fatal especially in kittens [75] . The virus is transiently excreted in feces, urine, saliva and vomiting and its high resistance in the environment (still infectious after 13 months at 4-25uC; [76] ) makes indirect transmission through feces and contaminated areas largely predominant [77, 78] . 3.3. Serological data. The serological statuses for FIV, FHV, FCV and FPV were obtained in 2007 in 15 natural rural populations of domestic cats in North-Eastern France [62, 79] . Cats were captured using baited traps or directly caught by the owner, anaesthetized, measured, and blood samples were taken from the jugular vein. FIV-antibodies were immediately searched for with a commercial kit using the ELISA method (SNAP Combo +, Idexx), whereas specific antibodies against FHV, FCV or FPV were measured by a specific blocking ELISA [80] . None of the cats was vaccinated. All six pairs of viruses were tested for potential association. Between 467 and 474 cats were tested for each virus and 465 to 469 were double-tested (depending on the virus pair). Previous analyses using logistic regression models with the same dataset revealed the combination of risk factors that were supported by our data [62] . Five factors were initially investigated: age (AGE), sex (SEX), way of life (owned or unowned, WOL), orange phenotype (orange or non orange, PHENO) and body mass (MASS) and one correction factor (the population of origin, POP) was considered. For each virus, the most appropriate model was selected using the Akaike Information Criterion adjusted for small sample size (AICc, [81] ). Ideally, all factors potentially creating apparent associations should be included in the model. But to limit the number of correction risk factors, the minimal model containing the identified risk factors for the two viruses was retained as a compromise for each pair (Table 1) . The corrected chi-square was robust for all tested sample sizes and numbers of parameters, whatever the nature of the factors (scenarios i, ii, iii) and the method used to calculate the P-value (P-value1, P-value2, see File S1 and Fig. S3 for more details). The type I error of this method remained indeed very close to 5% (Fig. 1) . On the contrary, the robustness of the logistic regression approach decreased with the N F /n ratio (number of factors/sample size). In scenarios i (only qualitative factors) and ii (only quantitative factors), the type I error was around 5% for a ratio of 0.005, around 8% for a ratio of 0.15 and around 20% for a ratio of 0.35. It became significantly different from 5% for ratios larger than 0.12 (type I error = 6.7%, z = 2.47, p = 0.019) and 0.08 (type I error = 7.9%, z = 4.21, p = 5.7610 25 ) for scenarios i and ii, respectively. In the mixed scenario (iii), the type I error became significantly different to 5% for all N F /n ratio larger than 0.075 (type I error = 7.1%, z = 3.047, p = 0.0038). More details are available in File S1 and Fig. S2 . Taken together, these results show that, as a rule of thumb, the logistic regression approach is robust for N F /n ratios below 0.1 for all types of factors. The two approaches (corrected chi-square and logistic regression) were used for the analysis of the interactions between four cat viruses ( Table 2) . Results showed that the interaction was not significant for pairs involving FIV. All other pairs (FHV-FCV, FHV-FPV and FCV-FPV) were found to interact, i.e., the number of individuals coinfected by two viruses could not be explained by shared risk factors. The three significant associations were all positive, meaning that there were always more co-infected individuals than expected considering shared risk factors (Table 3) . Pairwise interactions between FHV, FCV and FPV could have come from the fact that one virus was a common risk factor for the two others. This possibility was tested (see the three last lines of Table 2 ) by adding the serological status to one virus as a common risk factor for the two others. Results led to reject this hypothesis, meaning that the observed associations cannot be solely explained by the fact that one virus interacts with the two others. The two P-values obtained for the corrected chi-squares are coherent. As for the P-values obtained for the logistic regression approach, they are usually slightly lower than those of the corrected chi-squares, probably because of the over-predictive trend of logistic regressions. In addition, as with simulated data, the logistic regression approach was less robust to small sample sizes than the corrected chi-square (Table S1 ). This was tested by randomly sampling smaller subsets of the cat data in order to increase the N F /n ratio. Finally, to emphasise the need to consider risk factors in the analysis of interactions, we also calculated the classical independence Pearson's chi-square. This approach, which does not integrate risk factors, predicted an association between five of the six tested pairs. In the case of the FIV-FCV and FIV-FHV pairs, it would lead to wrongly conclude on the existence of an interaction, whereas the two approaches have shown that these apparent interactions were in fact explicable by shared factors. Common risk factors can create statistical associations. This work confirmed that ignoring them would lead to wrong conclusions. Ignoring them would indeed result in an over-estimation of the number of interactions as any association, biological or statistical, would be put in one basket. The loss of significance after controlling for other factors was illustrated in this paper with feline viruses data, and was previously found by Behnke et al. [39] for helminth parasites of the wood mice. The next step was to identify an appropriate way to take those risk factors into account. Two approaches to take risk factors into account with serological data (i.e., presence-absence) were proposed and examined. Those are the use of logistic regression models as Table 1 . Risk factors models used to test for potential association between pairs of feline viruses. previously done by some authors [52, 57] , or an adaptation of the chi-square test for independence presented for the first time in this paper. To determine which method should be used under which circumstances, we need to make the following considerations. First, the corrected chi-square involves 2n+2 estimations of the logistic regression coefficients, n being the number of bootstraps. In comparison, only two models must be parameterized in the logistic regression. As a consequence, the logistic regression approach is much faster to run (less than a second versus 2.5 minutes for the corrected chi-square for a model with 6 factors in full interaction and 300 individuals, for 1000 bootstraps, using a desktop computer with an Intel(R) core(TM)2 Quad CPU Q6600 processor). Second, the corrected chi-square is more robust than the logistic regression, especially for small sample size. A first solution would be to use the corrected chi-square as soon as simulation times are acceptable. For a 5% rejection threshold, a more straightforward alternative is to use the corrected chi-square by default as soon as the ratio between the sample size and the number of parameters is below 10 and the logistic regression in the opposite case. However, we did not test all potential situations and further analyses are needed to determine the limit of robustness of the logistic regression approach (in particular in situations where the probability of infection is not 50% and can be affected by risk factors). Two P-values have been proposed for the corrected chi-square. The first one relies on the assumption that the corrected chi-square is proportional to a chi-square with one degree of freedom; the second one simply counts the proportion of in silico datasets for which the value of the corrected chi-square is above the observed value. Both P-values led to consistent results using a 5% rejection threshold, consistently with the fact that for all tested pairs the corrected chi-square fitted well with an under-dispersed chi-square with one degree of freedom (Fig. S1, Fig. S2 ). Which one should be used in practice actually depends on the simulation time. If simulations are fast enough and if running 1000 bootstrap is acceptable, P-value2 should be preferred. In the opposite case, a good option is to run much less bootstraps (typically 30) and to use P-value1. Even if other alternative methods allow taking covariates into account, we only compared the corrected chi-square to the logistic regression approach. We could have compared it as well to loglinear models, which model the probability of infection with single and multiple parasite species from contingency tables and allow including known risk factors. However, in this approach the independence between parasites is tested using likelihood ratio tests, which are based on an asymptotic approximation of the deviance as in the logistic regression approach. They should therefore have the same limitations than logistic regressions and their robustness should be similarly influenced by the N F /n ratio. In addition, continuous variables are usually discretized in loglinear models, whereas the corrected chi-square allows working with continuous data. After correction by the known risk factors of the viruses, three pairs of feline viruses out of six appeared to be significantly associated. The N F /n ratio being 0.04 to 0.06, the logistic regression approach can be considered robust, at least for a 5% rejection threshold. First, it is worth noting that age is a crucial covariate. The infection probability of all viruses increases with host' age [62] , thus age must strongly participate in the generation of false interactions. This age-dependence is due to both a biological effect (i.e., behaviors and immune defenses may evolve with age, [82, 83] ) and a mechanical effect (i.e., older individuals are more likely to be seropositive because of a longer exposure time). Disentangling both effects would require the use of Susceptible-Infected-Recovered (SIR) models, but was not necessary here. Indeed, to The type I error of the corrected chi-square tests represented here is based on P-value2 but similar results were observed with P-value1 (Fig. S3) . Note that for the logistic regression approach, points resulting from a given sample size were linked to see the effect of the N F /n ratio for different sample sizes (solid line: n = 100, dashed line: n = 200, dotted line: n = 300). The dashed horizontal line represents a type I error of 5%. doi:10.1371/journal.pone.0029618.g001 correct for age in the study of interactions, the important is to model the evolution of the probability of infection with age. Correcting for all risk factors, no pair of viruses involving the Feline Immunodeficiency Virus (FIV-FHV, FIV-FCV, FIV-FPV) was significantly associated. This result is at first surprising because, as in humans infected by HIV, feline AIDS is characterised by a chronic immunodeficiency, allowing subsequent opportunistic infections (review in [84] ). Indeed, although FIV positive cats can mount immune responses to administered antigens other than during the terminal phase of infection, their primary immune responses may be delayed or diminished [85, 86] . Experimental studies also revealed that cats co-infected by FIV and FCV or FHV had more severe disease signs than non-FIV infected cats [87, 88] . In addition, the presence of FHV was shown to accelerate FIV transcription through the activation of the FIV long terminal repeat [89] , a phenomenon that was also shown in vitro for the human versions of the viruses, HSV2 and HIV [90] [91] [92] [93] . Those laboratory experiments show that FIV infection may increase the severity of FHV or FCV-induced clinical signs but do not address the question of the effect of FIV on the sensitivity to FHV or FCV infection. Furthermore, the few epidemiological studies interested in the question did not demonstrate any epidemiological association between FIV and FHV [94] . In other words, if experimental investigations suggest a synergy between FIV and FHV and between FIV and FCV towards a more severe disease, our sero-epidemiological study suggests that the identified risk factors explain by themselves the apparent increase of double sero-positive individuals. As for the FIV-FPV pair, this study is to our knowledge the first to search for a potential association. Whether risk factors were taken into account or not, we did not find any significant association between the two viruses. Again, this could be at first surprising as both viruses are supposed to be immunosuppressive [84, 95, 96] . In experimental conditions, FPV infection is more severe in FIV-infected cats [97] . Consequently, a positive association could have been expected if infections had facilitated each other (leading to numerous co-infections) or a negative association if the co-infection had led to a strong host mortality (leading to few co-infections). However, the FPV-induced decrease in the immune response is transient and more likely to occur in young kittens, whereas FIV infection is more frequent in adult cats. The persistence of FPV-antibodies can be longer than 7 years [98] , and consequently, double seropositivity against FPV and FIV is not synonymous of co-infection. It is likely that co-infections by the two viruses are not frequent and mainly occur in adult animals which are less sensitive to FPV. As no association was evidenced for these three pairs of viruses, the FIV infection does not seem to modify the risk of infection by another virus. However, our results do not exclude the occurrence of an interaction once both parasites are in contact within the host (e.g., directly through competition or indirectly via the host immune system), as suggested by several experimental co-infection studies. In addition, the FIV seropositivity status may encompass different stages of the infection with various degrees of immunodeficiency. The results of this study do not exclude the possibility that late stage FIV infection may increase the sensitivity to the other feline viruses. On the contrary, the three other pairs (FHV-FCV, FHV-FPV and FCV-FPV) were significantly associated after correction by their known risk factors. It is to our knowledge the first evidence of a possible interaction between those viruses. As more double seropositive cats than expected under the independence hypothesis were observed, possible synergies are suggested. After an acute infection, FHV is known to persist life-long in a latent form, which can be reactivated in stressful conditions [73] . Infection with FPV or FCV could thus be responsible for the reactivation of FHV in latently infected animals, resulting in seroconversion against both FHV and the new infecting virus. This could explain the FHV-FCV and FHV-FPV associations. In addition, since FPV is more immunosuppressive than FCV, the interaction between FPV and FHV is expected to be stronger than that between FCV and FHV, which is consistent with our results. The immunosuppressive effect of FPV could also explain the association with FCV. In that case however, contrary to FHV, it would require that the FCVinfection occurs at the time of the immunosuppression occuring within the two weeks post-FPV infection. Interestingly, a similar association between FPV and FCV antibodies was described in free-ranging lions in East Africa [99] . This work pointed out new probable synergies between feline viruses that can now be further investigated in laboratory conditions. However, the associations could also result from the existence of an unknown confounding factor common to FHV, FCV and FPV. The feline parvovirus is immunosuppressive, as a result of the strong leukopenia occurring within the two weeks post-infection [95, 96] . This virus could therefore be a confounding factor to the FHV-FCV pair if FPV-seropositive cats are more susceptible to FHV and FCV at the same time. However, as shown in this paper, the FHV-FCV interaction remained significant after correction by FPV (Table 2) . If FPV is not a confounding factor, we cannot exclude the existence of another one, such as a greater susceptibility of certain individuals to infections whatever the parasite involved. Numerous studies have shown that an extensive inter-individual variability exists in response to certain pathogens, such as HIV (review in [100] ), trypanosomiasis (review in [101] ), or human and bovine tuberculosis (reviews in [102, 103] ), including variations in susceptibility to the parasite, its transmission, and/or the course of disease progression. It has been attributed to host determinants and variability in multiple genes that regulate virus cell entry, acquired and innate immunity (e.g., macrophages, molecular and cellular actors of the inflammatory reaction), and others that influence the outcome of the infection. Hosts with a diminished or delayed innate immune response may in fact be more susceptible to any infection, with physiological parameters, such as hormonal profiles (e.g., [104] ), possibly playing a role in the modulation of transmission efficiency and/or in the immune response intensity. A weaker physical condition could also lead to a higher sensitivity to infectious agents (lower dose-effect, different intra-host dynamic) (e.g., [105] ). More generally, individuals' personality may as well be involved [61, 106] . A better understanding of genetic, physiological and immunological basis of such inter-individual variability would therefore be of particular interest in the context of polyparasitism. Another perspective of this work is the development of new methods able to distinguish pairwise interactions from those due to common confounding factors shared by the three viruses. Such methods could use the proportion of infected individuals that are in reality triply infected. While the study of macroparasites usually uses quantitative data (i.e., parasite load per individual host), the study of microparasites on the field is most of the time limited to presence-absence data (i.e., serology), making the detection of associations between parasites more complicated from a methodological point of view. The corrected chi-square proposed in this study is, with the logistic regression approach, currently one of the rare ways to search for interaction between parasites from presence-absence data. This work provides evidence of the efficiency of such methods to reduce the bias introduced by common risk factors and encourages their use. However it also points out the low robustness of the likelihood ratio test for certain data characteristics. The corrected chi-square test must indeed be preferred for small sample size. Those methods can be applied to any epidemiological study based on serology, within human or animal host populations. Applied here to feline viruses, they revealed significant associations between three pairs of feline viruses. If they still do not allow us to decide whether such associations are really true interactions or whether they reveal the existence of ''over-susceptible'' hosts, we believe it is an important step forward as it offers the possibility to point out parasites associations that should be further investigated in experimental conditions. The understanding of parasites interactions and of their consequences on diseases evolution, emergence and management is indeed a crucial challenge for human and animal epidemiologists of our time. Figure S2 Issue of the conformity tests of the type I error to 5% according to the N F /n ratio for the logistic regression approach. The issue was coded 1 when the test was significant, 0 when not and the resulting logistic regression was drawn (dark line). Three scenarios are considered: i) all factors are qualitative (A); ii) all factors are quantitative (B) and iii) half of the factors are quantitative and the other half are qualitative (mixed scenario, C). (EPS) Figure S3 Type I error (%) of the corrected chi-square tests according to the N F /n ratio and the type of P-value used for the corrected chi-square: P-value1 (blue empty points) or P-value2 (red full points). Three scenarios are considered: i) all factors are qualitative (A); ii) all factors are quantitative (B) and iii) a half of the factors is quantitative and the other half is qualitative (mixed scenario, C). The dashed horizontal line represents a type I error of 5%. Table S1 Corrected chi-square tests and logistic regressions to search for feline viruses' interactions using subsets randomly sampled in cat data such that the N F / n ratio takes various values. File S1 Robustness of the logistic regression approach and of the corrected chi-square test. (1) Conformity tests of the type I error to 5%, (2) Influence of the way to calculate the Pvalue of the corrected chi-square test on the robustness of the study. (DOC) File S2 ''Chi2corr'', an R program for the application of the corrected chi-square test to any presence-absence data: test statistic, observed and expected frequencies, estimated dispersion coefficient (parametric bootstrap), P-values and distribution of the bootstrapped corrected chi-square. File S3 A step-by-step example of application of the corrected chi-square test to search for interaction between two parasites, using a provided dataset (''da-ta_example.txt'', File S4) and the provided R program (''Chi2corr.R'', File S2).

The Transmembrane Domain of CEACAM1-4S Is a Determinant of Anchorage Independent Growth and Tumorigenicity

CEACAM1 is a multifunctional Ig-like cell adhesion molecule expressed by epithelial cells in many organs. CEACAM1-4L and CEACAM1-4S, two isoforms produced by differential splicing, are predominant in rat liver. Previous work has shown that downregulation of both isoforms occurs in rat hepatocellular carcinomas. Here, we have isolated an anchorage dependent clone, designated 253T-NT that does not express detectable levels of CEACAM1. Stable transfection of 253-NT cells with a wild type CEACAM1-4S expression vector induced an anchorage independent growth in vitro and a tumorigenic phenotype in vivo. These phenotypes were used as quantifiable end points to examine the functionality of the CEACAM1-4S transmembrane domain. Examination of the CEACAM1 transmembrane domain showed N-terminal GXXXG dimerization sequences and C-terminal tyrosine residues shown in related studies to stabilize transmembrane domain helix-helix interactions. To examine the effects of transmembrane domain mutations, 253-NT cells were transfected with transmembrane domain mutants carrying glycine to leucine or tyrosine to valine substitutions. Results showed that mutation of transmembrane tyrosine residues greatly enhanced growth in vitro and in vivo. Mutation of transmembrane dimerization motifs, in contrast, significantly reduced anchorage independent growth and tumorigenicity. 253-NT cells expressing CEACAM1-4S with both glycine to leucine and tyrosine to valine mutations displayed the growth-enhanced phenotype of tyrosine mutants. The dramatic effect of transmembrane domain mutations constitutes strong evidence that the transmembrane domain is an important determinant of CEACAM1-4S functionality and most likely by other proteins with transmembrane domains containing dimerization sequences and/or C-terminal tyrosine residues.

CEACAM1 is a member of the carcinoembryonic antigen (CEA) gene family of Ig-like cell-cell adhesion molecules [1, 2] . Like other members of this family, CEACAM1 is a type I transmembrane protein with a heavily glycosylated extracellular region composed of four Ig-like domains, a transmembrane domain and a cytoplasmic tail [3] . In the rat liver there are two allelic variants of CEACAM1 which differ by 16 amino acids in their N-terminal domains [2, 4] and two major splice variants, designated 4L and 4S, that are distinguished by differences in the length of their cytoplasmic tails of 70-72 amino acids and 10-12 amino acids, respectively [2, 4, 5] . Both isoforms of CEACAM1 are down-regulated in epithelial cancers arising in the liver, prostate, bladder and colon [6, 7, 8, 9, 10, 11, 12] , a finding that prompted re-expression analysis aimed at defining structure and function relationships. Restoration of expression by infecting rat cell lines derived from primary hepatocellular carcinomas (r-HCC) with a CEACAM1-4L retrovirus resulted in potent growth suppression in vitro and tumor suppression in vivo [13] . Further analysis showed that the 4L cytoplasmic domain was necessary and sufficient for tumor suppression [14] , an activity that required phosphorylation of serine 503 and in colon carcinoma cells, concurrent phosphorylation of tyrosine 488 [15, 16] . In contrast to CEACAM1-4L, CEACAM1-4S failed to generate a tumor suppressor phenotype when re-expressed in r-HCC or mouse colon carcinoma cell lines [13, 17, 18] . However when expressed in MCF7 mouse mammary carcinoma cells, CEA-CAM1-4S induced glandular morphogenesis, an activity requiring phosphorylation at one or more sites in the 4S cytoplasmic domain [19, 20, 21] . Site directed mutagenesis further showed that mutation of phenylalanine 445 at the C-terminus of the CEACAM1-4S cytoplasmic domain not only compromised interactions with the actin cytoskeleton but also inhibited lumen formation, suggesting interactions of CEACAM1-4S with the cytoskeleton were an important determinant of glandular morphogenesis. Interestingly, when mouse mammary carcinoma cells were grown in humanized NOD/SCID mouse mammary fat pads, only the 4L isoform initiated morphogenesis, the opposite of what was observed in vitro [21] , raising questions about the equivalence of in vitro and in vivo models of morphogenesis. Because of its role in cell adhesion, the CEACAM1 N-terminal Ig domain [22, 23, 24] , like the cytoplasmic domain, has been the focus of numerous investigations. The adhesive epitope within the N-terminal Ig-domain has been defined for rat [24] , mouse and human CEACAM1 [22, 23] , the evolutionary relationships between CEACAM1 from different species has been determined [25, 26] and the three dimensional structure has been established by X-ray crystallography [27] . In comparison, the CEACAM1 transmembrane domain has received relatively little attention, perhaps because transmembrane domains have often been viewed as passive anchor sequences that span the lipid bilayer. Over the last 10 years, this simplistic viewpoint has fallen by the wayside in the face of accumulating evidence implicating transmembrane domains in helix-helix interactions leading to dimerization, oligomerization and signal transduction [28, 29, 30] . The possible involvement of transmembrane-transmembrane domain interactions in the functionality of CEACAM1 was suggested by the presence of repeating GXXXG motifs (where X represents any amino acid), sequences known to control protein dimerization and signaling [30, 31] , and the presence of transmembrane C-terminal tyrosine residues shown in other proteins to be mediators of molecular recognition, self assembly and signal transduction [32] . In the present investigation, we have examined the effect of transmembrane domain mutations on the ability of CEACAM1-4S to confer an anchorage independent phenotype when expressed in a clonal line of CEACAM1 negative, anchorage dependent rat hepatocellular carcinoma cells, designated 253-NT. Our results show that transmembrane mutations in both GXXXG and tyrosine residues have both positive and negative effects on the anchorage independent phenotype produced by wild type CEACAM1-4S. The origin and characteristics of MAb 5.4 specific for CEACAM1 and MAb 188.A2 specific for rat transferrin receptor have been described previously [33, 34] . Monoclonal antibody 9.2 (MAb 9.2) was provided by Drs. Werner Reutter and Oliver Baum at the Free University, Berlin, Germany [35] . Mouse anti-human HLA antibody was purchased from Sigma-Aldrich (Sigma-Aldrich Co., St. Louis, MO). The preparation of polyclonal rabbit antipeptide antibodies specific for the CEACAM1-4L and CEA-CAM1-4S has been previously described [36] . The secondary antibodies used for indirect immunofluorescence labeling were Alexa-488 conjugated goat anti-mouse and goat anti-mouse-HRP conjugated secondary antibody (Invitrogen, Carlsbad, CA, USA). The parental cell line 253T was established from a 2acetylaminofluorene induced rat hepatocellular carcinoma, as described previously [35] . The anchorage dependent 253T-NT cell line was isolated from 253T by limiting dilution cloning. 253T and 253-NT cells were grown in Waymouth medium (Sigma, St. Louis, MO, USA) supplemented with 15% FBS, 1% glutamine (Invitrogen), 0.1% Gentamycin (Invitrogen), and 0.2% Normocin (Invivogen, San Diego, CA, USA). For cell proliferation assays, 1.5610 4 cells were plated in a 24-well plate. At 24, 48, 72, and 96 hours after plating, cells were trypsinized (Invitrogen), stained with trypan blue and counted using a hemocytometer. RNA was isolated from a normal Fisher rat liver using RNAzol B according to the manufacturer's instructions (Tel-Test, Friendswood, TX). cDNA was synthesized from the purified RNA according to the manufacturer's instructions using the SuperScript III first-strand synthesis system for RT-PCR (Invitrogen). CEACAM1-4S cDNA was amplified by PCR from the total cDNA product using primers: CEACAM1-4S Forward 59CAGGAATT-CATGGAGCTAGCC-TCGGCT-39 and CEACAM1-4S Reverse 59-CGAGTCGACT-CGTCAGAAGGAC CCAGATCC-39. The primers contained EcoRI and SalI restriction sites. A restriction digest was performed on both the CEACAM1-4S PCR product and the pCl-neo plasmid (Promega, Madison, WI, USA) using EcoRI and SalI (New England Biolabs, Ipswich, MA, USA). Following digestion, the plasmid and the PCR product were dephosphorylated using Antarctic phosphatase (New England Biolabs), heat-treated to inactivate the phosphatase, and run on a 1% agarose gel. Bands corresponding to the plasmid DNA and CEACAM1-4S PCR product were purified using the Geneclean spin kit (Qbiogene, Morgan Irvine, CA, USA). The CEACAM1-4S PCR product was ligated into the pCI-neo plasmid using T4 DNA ligase (New England Biolabs). Ligated plasmid was transformed into One Shot OmniMax 2 T1 Phage-Resistant Cells (Invitrogen) according to the manufacturer's protocol. Transformed cells were plated onto LB/ CARB plates and resulting colonies were screened by PCR using forward and reverse primers for CEACAM1-4S (see above). Plasmid DNA from four CEACAM1-4S positive clones was isolated using the Qiagen Endofree maxiprep kit (Valencia, CA, USA) and submitted for DNA sequencing to the W.M. Keck Biotechnology Laboratory at Yale University (New Haven, CT). Mutation sites were chosen based on the consensus location of the transmembrane domain predicted by the different membrane topology algorithms shown in Table 1 . The results from this analysis showed that the various transmembrane prediction algorithms delineated transmembrane domains with variable N (G424 to I431) and C (Y445 to S449) termini and different numbers of potential GXXXG motifs. All of the predicted transmembrane domains lacked the terminal GLSE predicted by Kyte Doolittle and Hopp-Wood (Table 1 ) and thus did not include the G420-LSE-G424 motif. Ten of the 12 predicted transmembrane domains included G428, strongly suggesting the presence of the two motifs centered on G432 (G428-IVI-G432 and G432-SVA-G436). Although G424 was present in only 3 of 12 predicted TM, G424's location at the N-terminus of a tandem array of three classic GXXXG motifs caused us to consider that the G424-AIA-G432 sequence might function as a dimerization motif when all the other GXXXG motifs were disrupted by G428L and G436L mutations. With regard to the C terminus, all of the predicted transmembrane domains contained Y445 and 7 of 12 contained Y448, leading us to conclude with reasonable certainty that Y448 was within the transmembrane domain or at the very least, at the interface between the transmembrane and cytoplasmic domains. Taking all of these considerations into account, we arrived at the transmembrane sequence proposed in Table 2 . The brackets denote the uncertainty with regard to G424 and Y448. In recognition of the variability in the predicted N-terminus and the uncertainty regarding the functionality of the G424-AIA-G428 motif, we introduced a G424L mutation to disrupt this motif and a G432L mutation to knock out both the G428-IVI-G432 and the G432-SVA-G436 motifs. Disruption of all three of the possible GXXXG motifs was accomplished by introducing both G424L and G432L mutations. To disrupt tyrosine mediated interactions, Y445 and Y448 were replaced with V, the choice of a tyrosine to valine mutation being based on the following considerations: 1) valine lacks a hydroxy group; 2) valine does not have an aromatic side chain; 3) valine is non-polar and thus should not disrupt the alpha helix or alter transport to the plasma membrane. Site-directed mutagenesis was performed on the pCI-neo plasmid containing the CEACAM1-4S gene using the Quik-Change II XL Site-directed mutagenesis kit (Stratagene, La Jolla, CA, USA) according to the manufacturer's protocol. The oligonucleotide primers used for mutagenesis were as follows: For transfection of 253-NT cells, 1610 6 cells at 75% confluence were transfected with 7.5 mg of plasmid DNA in Lipofectamine LTX reagent (Invitrogen) following the manufacturer's recommended protocol. After incubation for 24 hours in complete medium, stable transfectants were selected by maintaining cells in complete medium containing geneticin (Invitrogen) at 600 mg/ml with medium changes every 48 hours. For all subsequent experiments, geneticin resistant cells were maintained in complete selection medium. Cells were seeded in permanox chamber slides (Nalge Nunc International, Rochester, NY) at a density of 1610 5 cells/ml and incubated in a 5% CO 2 humidified chamber for 48-72 hours. Cells were washed three times in ice-cold PBS and fixed for 10 min in ice-cold acetone. Normal liver sections and frozen tumor sections were prepared as previously described [37] . Cells were blocked for 15 min with 1% normal goat serum (Sigma-Aldrich) in PBS and incubated for 30 min at room temperature with a 1:200 dilution of primary MAb 9.2 specific for CEACAM1-4S. After three washes in PBS, cultures were fixed with 4% paraformaldehyde for 1 min, quenched in 0.1 M glycine in PBS, and incubated at 4uC for 30 min in Alexa 488 conjugated goat anti-mouse IgG (1:400 dilution). Cells were examined by fluorescence microscopy using a Nikon Eclipse E800 microscope (Nikon Instruments, Inc., Melville, NY). Confocal images were acquired with a Nikon PCM 2000 (Nikon Inc. Mellville, NY, USA) using the Argon (488) and the green Helium-Neon (543) lasers. Serial optical sectioning was performed with Simple 32, C-imaging computer software (Compix Inc, Cranberry Township, PA, USA). Z series sections were collected at 0.5 mm using a 606 PlanApo objective and a scan zoom of 26. Images were processed in NIH Image J (National Institutes of Health, Springfield, VA, USA). Cells were harvested at 75% confluence using non-enzymatic cell dissociation solution (Sigma-Aldrich) and labeled in suspension with anti-CEACAM1-4S monoclonal antibody MAb 9.2 as described above. Primary and secondary antibodies were diluted in 1% normal goat serum in PBS and incubated sequentially with cells at 4uC for 20 min. Cells were washed with sterile ice cold sort buffer (Ca/Mg++ free, pH 7.0 PBS, 5 mM EDTA, 2 mM HEPES and 1% FBS) and resuspended at a final concentration of 8-10610 6 cells/ml. CEACAM1-4S positive cells were isolated by FACS as previously described by Comegys et al. [38] . Cells were harvested using non-enzymatic cell dissociation solution (Sigma-Aldrich), suspended in 0.76% low melting point agar diluted 1:1 with 26 complete Waymouth medium (Sigma-Aldrich) and seeded in 6 well plates (10 4 cells/well) coated with 2 mls of 1.25% low melting point agar (Sigma-Aldrich) diluted 1:1 with 26 complete Waymouth medium. Plates were incubated at 37uC for three weeks and analyzed microscopically to identify anchorage independent clones. Colony size was determined from digital micrographs using Image-Pro Plus 5.0 software (Media Cybernetics, Inc., Bethesda, MD) to determine the average area calculated from measurements made on a total of 15-30 colonies. All animal studies were performed using protocols approved by the Rhode Island Hospital Institutional Animal Care and Use Committee (cmtt# 0268-98). Athymic nude mice were purchased from Harlan (Indianapolis, IN). Twenty-four hours prior to transplantation of tumor cells, animals were injected intraperitoneally with anti-LY2.2 antibodies containing anti-asialo GM1 to suppress T-cell and NK cytotoxic activity, respectively [39, 40] . Cultured cells were harvested as described previously [13] . Subcutaneous injections with 5610 6 cells per site were performed under aseptic conditions into both of the upper flanks. For intraperitoneal injections, mice were injected with 2610 6 cells suspended in HBSS. Tumors were excised at three weeks postinjection, weighed, frozen in a hexane/acetone bath and stored at 280uC. Tumors were analyzed by indirect immunofluorescence as described above. Immunoblot Analysis Following Blue-Native (BN) PAGE BN-PAGE analysis was performed on crude membrane isolates prepared from 253T-NT cells stably transfected with empty vector, wild type CEACAM1-4S and each of the single and double glycine mutants described above. Cell pellets were thawed on ice and lysed following 10 passages through a 20-gauge needle. Lysates were centrifuged at 3,500 rpm for 15 min at 4uC and MgCl 2 and benzonase (Sigma Aldrich) were added to the supernatant to give final concentrations of 2 mM and 1 unit/ml, respectively. After incubation at room temperature for 30 min, membranes were centrifuged at 17,000 rpm for 15 min at 4uC. Blue-native (BN) PAGE electrophoresis was performed using the Native Page Novex Bis-Tris gel system (Invitrogen). Membrane pellets were solubilized on ice for 30 min in 16 Native Page sample buffer (Invitrogen) containing 1% dodecyl-ß-D-maltoside (Invitrogen) and protease inhibitors. After a 30 min incubation, the samples were centrifuged at 20,0006 g for 20 min at 4uC to remove insoluble material. A Bradford assay (Bio-Rad, Hercules, CA, USA) was performed to determine protein concentration. One ml of 5% NativePage G-250 sample additive (Invitrogen) was added to each sample, and the samples and a Native Mark protein ladder (Invitrogen) were loaded into the wells of a 4-16% Bis-Tris gel (Invitrogen). After running at 150 V for approximately 30 min, the cathode buffer was changed from 0.02% to 0.002% G-250 and electrophoresis at 150 V was continued for an additional 120 min. BN-PAGE gels were immunoblotted onto PVDF membranes (Bio-Rad). Transfer was performed using Bio-Rad's semi-dry transfer apparatus for 1 hour at 25 V. After transfer, membranes were incubated in 8% acetic acid for 15 min, rinsed with water and air-dried. PVDF membranes were re-hydrated with methanol, rinsed with water and blocked in 5% milk overnight at 4uC. Blots were labeled with antibodies as previously described [24] . Immunoblots were prepared with extracts from 80% confluent cell cultures lysed in RIPA buffer (Pierce, Rockford, IL, USA) containing protease and phosphates inhibitors (Calbiochem). Cell lysates were centrifuged at 14,000 rpm for 15 min and resolved by SDS-PAGE. Immunoblots were prepared and visualized as previously described, using MAb 9.2 to detect CEACAM1-4S [13] . The density of the MAb 9.2 reactive bands was determined by image analysis of digital images of immunoblots captured using a Versadoc Imaging System (Bio-Rad) and Quantity One software (Bio-Rad), as previously described [13] . A paired t-test was performed to determine statistical significance using GraphPad QuickCalcs software (GraphPad Software Inc., La Jolla, CA). In previous investigations, we focused on the structural and functional determinants of tumor suppression mediated by CEACAM1-4L, one of two major splice variants expressed in rat liver [13, 16] . The primary goal in these studies was to gain insight into the role of the cytoplasmic and extracellular domains in cell adhesion and tumor suppression. In the present report, we have focused on the CEACAM1 transmembrane domain, a relatively uncharacterized region shared by both the 4L and 4S isoforms. Our cell model for these studies was an anchorage dependent, CEACAM1 negative subclone designated 253-NT that was derived clonally from the parental rat 253T cell line [6] . When suspended in soft agar, 253T cells continued to grow and by three weeks had formed well-defined anchorage independent colonies ( Figure 1A ). In comparison, soft agar cultures of 253-NT cells showed no evidence of significant growth and after three weeks were composed almost entirely of single cells ( Figure 1B) . In contrast, 253-NT cells stably transfected with a wild type CEACAM1-4S expression plasmid and enriched by FACS to yield cultures at least 70% positive for CEACAM1-4S expression regained the anchorage independent phenotype of the parental 253T cells ( Figure 1C ) while cells stably transfected with empty vector ( Figure 1D ) had not grown significantly after 3 weeks in soft agar. Wild Type and Transmembrane Domain Mutants were Expressed at Similar Levels on the Surface of Stably Transfected 253T-NT Cells Sequence analysis showed that the CEACAM1-4S transmembrane domain contained multiple GXXXG dimerization motifs and two C-terminal tyrosine residues which in other transmembrane receptors were involved in transmembrane domain dependent signaling events [30, 31, 32] . To determine if these transmembrane domain elements played a part in the anchorage independent phenotype induced by CEACAM1-4S, nine expression vectors encoding CEACAM1-4S with transmembrane domain mutations in the GXXXG motifs and/or tyrosine residues were constructed as described under Methods and shown in Table 1 . These vectors were used to produce stably transfected sublines of 253-NT. Single G424L or G432L mutations or double G424L/G432L mutations were introduced to disrupt two or more GXXXG motifs. Substituting leucine for glycine introduced a large hydrophobic amino acid that maintained the hydrophobic character of the transmembrane domain but disrupted the conformation of the transmembrane domain helix [28] . Transmembrane domain tyrosine residues were mutated to valine, a substitution that replaced tyrosine with a non-polar hydrophobic amino acid lacking a hydroxyl group in its side chain. In other systems, proper spacing of the tyrosine hydroxyl group had been shown to be necessary for signaling events dependent upon protein-protein interactions [41] . Examination by confocal fluorescence microscopy of cultures established from CEACAM1-4S positive cells isolated by FACS showed intense membrane fluorescence when labeled by indirect immunofluorescence with CEACAM1 specific MAb 9.2, confirming that both wild type and mutant forms of CEACAM1-4S were properly transported to the plasma membrane ( Figure 2 , A-L). Empty vector transfected and untransfected 253-NT cells showed no detectable reactivity with MAb 9.2 ( Figure 2 , E and J, respectively). Immunoblot analysis indicated that the wild type and mutant proteins all had an apparent molecular mass of 105 kDa, the expected size for CEACAM1-4S isoform [2] , suggesting that post-translational processing has not been altered ( Figure 3 ). Quantitative analysis by flow cytometry of FACS selected, stably transfected cells labeled by IIF with MAb 9.2 indicated that 72-83% of the cells were positive for either wild type or mutated forms of CEACAM1-4S (Figure 4 ). The effect of transmembrane domain mutations on anchorage independent growth was determined from changes in the average areas of soft agar colonies ( Figure 5 and 6) . When 253T-NT cells expressing wild type ( Figure 1C) , single or double tyrosine mutants of CEACAM1-4S were grown in soft agar, sublines expressing tyrosine mutants (Figure 5 D, E, F and Figure 6 ) showed a 3.5-fold increase in colony size relative to cells transfected with the wild type protein ( Figure 1C and Figure 6 ). In contrast, 253T-NT cells expressing CEACAM1-4S with GXXXG motifs disrupted by G to L mutations ( Figure 5A , B, C and Figure 6 ) either did not grow or formed colonies that for the double glycine mutant ( Figure 5C and 6), were less than half the size of those formed by cells expressing the wild type protein ( Figure 1C and Figure 6 ). Moreover, the rapid growth phenotype conferred by tyrosine to valine mutants appeared to be dominant over the growth suppressed phenotype displayed by glycine to leucine mutants since all of the sublines with both glycine to leucine and tyrosine to valine mutations displayed enhanced anchorage independent growth ( Figure 5G , H, I and Figure 6 ). Proliferation assays were carried out to ascertain whether the size of soft agar colonies ( Figure 5 and 6 ) was proportional to the rate of proliferation in vitro calculated from changes in cell number as a function of time. As shown in Figure 7 , 253T-NT cells expressing wild type CEACAM1-4S proliferated 2.2 times faster than 253T-NT cells carrying the empty vector and from 1.4-2.2 times faster than cells transfected with glycine mutants. Consistent with their rapid growth in soft agar, 253T-NT cells expressing CEACAM1-4S with single or double tyrosine mutants proliferated at rates that were 1.4-1.6-fold higher than cells with wild type CEACAM1-4S. Taken together, these data suggested that the CEACAM1-4S transmembrane domain was controlling interactions involved in growth under anchorage independent conditions, interactions that were altered by GXXXG or tyrosine mutations. To determine if changes in anchorage independent growth induced by transmembrane domain mutations were mirrored by altered tumorigenicity, nude mice were injected subcutaneously in the upper flanks with 253T-NT cell lines expressing wild type or mutant CEACAM1-4S. Subcutaneous tumors were harvested from animals at three weeks after injection, a time point chosen by necessity because of the large tumor burden in animals injected with 253T-NT cells expressing tyrosine mutants. On average, tumors produced by cells expressing wild type CEACAM1-4S were 5.5-and 6.5-fold larger by weight than those generated by cells transfected, respectively, with the empty vector or the single G424L mutant (Figure 8 ). While cells expressing the wild type protein formed tumors comparable in weight to those formed by cells expressing the double G to L mutant, 253T-NT cells expressing CEACAM1-4S with single Y448V, double Y445V/ Y448V or quadruple G424L/G432L/Y445V/Y448V mutations produced tumors that were 1.9-, 1.25-, and 2.14-fold larger than those produced by cells transfected with the wild type protein (Figure 8 ). Indirect immunofluorescence analysis of frozen tumor sections, confirmed that tumor nodules formed by 253-NT cells transfected with either wild type or transmembrane domain mutants remained strongly positive for CEACAM1-4S (Figure 9 ). Previous reports have demonstrated that dimerization via transmembrane domain helix-helix interactions are often mediated by GXXXG or GXXXA motifs within the transmembrane domain [30, 42] . To determine if the GXXXG motifs within the transmembrane domain of CEACAM1-4S played a role in the dimerization of CEACAM1-4S [36, 43] , the effect of G to L mutations on CEACAM1-4S interactions was analyzed by Blue-Native polyacrylamide gel electrophoresis (BN-PAGE). In BN-PAGE, Coomassie G-250 is used in place of SDS to coat proteins with a uniform negative charge without causing denaturation or disruption of protein-protein interactions [44] . As shown in Figure 10 , wild type and single G-to-L mutants of CEACAM1 demonstrated an apparent molecular mass by BN-PAGE that was approximately 100 kDa higher than the double G to L mutant, a difference approximately the size of CEACAM1-4S resolved by reducing SDS-PAGE. These data suggested that a single GXXXG motif was sufficient to mediate dimerization of CEACAM1-4S and/or interaction with another yet-to-be identified transmembrane protein, interactions that appeared to require helix-helix interactions mediated by GXXXG motifs. The ability of CEACAM1-4S expression in the context of the proteome of 253-NT cells to restore the tumorigenic and anchorage independent growth characteristics of the parental 253T cell line [6] provided a quantifiable, reproducible endpoint for examining the functionality of the CEACAM1-4S transmembrane domain. The idea that CEACAM1 phenotypes could be context specific came from our previous studies showing that CEACAM1-4L expression dramatically suppressed the tumorigenicity of CEACAM1 negative PC-3 human prostate carcinoma cells [13] . With continued passage, however, cells eventually reacquired a tumorigenic phenotype without losing expression of CEACAM1-4L, suggesting a strong selection for cells with proteomes that were unable to support CEACAM1 mediated tumor suppression. Context dependent effects on CEACAM1-4S phenotypes were also suggested by the ability of CEACAM1-4S to induce morphogenesis of MCF7 cells in vitro but not in vivo [21] . Analysis of the amino acid sequence of the CEACAM1-4S transmembrane domain revealed the presence of four GXXXG sequences (Table 1 and 2), a motif known to drive high affinity helix-helix interactions that stabilize the dimerization/oligomerization of many well characterized proteins such as glycophorin A, epidermal growth factor receptor and the G protein-coupled afactor receptor of budding yeast [30, 42, 45] . For the latter protein, disruption of the GXXXG motif not only impaired oligomerization but also disrupted signaling [30, 46] . Previous investigations have shown that both the long and short isoforms of CEACAM1 form cis-dimers [36, 43] , an interaction that we hypothesized should be stabilized by GXXXG mediated helix-helix associations. To test this idea, we introduced glycine-to-leucine mutations that disrupted one or more transmembrane domain GXXXG motifs. FACS, fluorescent confocal microscopic and immunoblot analysis of cells stably transfected with the glycine-to-leucine mutants confirmed that the mutant proteins were expressed on the cell surface at the same level and the same size as wild type CEACAM1-4S. However, after 3 weeks in soft agar, cells expressing single glycine mutants showed significantly lower rates of proliferation and significantly smaller colonies when compared to cells transfected with wild type CEACAM1-4S, suggesting that a single glycine-to-leucine mutation had compromised the ability of CEACAM1-4S to induce anchorage independent growth. From these results, we concluded that the smaller size of colonies produced by glycine mutants resulted at least in part from a decrease in the rate of proliferation. Since the double glycine mutation disrupted all of the GXXXG motifs, a corresponding decrease in either homo-or hetero-cisdimerization would be expected if these motifs were the sole mediators of this interaction. However, impaired dimerization could also occur if a reduction in GXXXG motifs destabilized cisdimers formed by single glycine mutants with a subsequent reduction in their steady state levels. Blue native-PAGE analysis showed that the apparent size of the double but not the single glycine mutants showed approximately a 100 kDA decrease in size, a shift consistent with impaired dimerization. Moreover, since single glycine to leucine mutations produced the growth suppressed phenotype but only double mutations hampered dimerization, it was concluded that even without an apparent effect on dimerization, glycine to leucine point mutations in the transmembrane domain had a profound effect on the growth phenotype produced by CEACAM1-4S. With one exception, the effects of transmembrane domain mutations on growth in vivo were similar to those observed in vitro. In keeping with their limited capacity for anchorage independent growth, cells transfected with empty vector were poorly tumorigenic relative to cells expressing the wild type CEACAM1-4S. As predicted from their growth in soft agar, cells expressing the G424L mutant formed tumors considerably smaller than those derived from cells expressing wild type CEACAM1-4S. However, the similarity in the size of tumors generated by cells expressing the wild type, double glycine and double tyrosine mutants was at odds with in vitro assays where cells expressing wild type CEACAM1-4S showed a significantly greater or lesser capacity, respectively, for anchorage independent growth than cells expressing double glycine or tyrosine mutants. This discrepancy is reminiscent of the differential in glandular morphogenesis exhibited in vitro and in vivo by MCF7 mammary carcinoma cells [21] and thus could reflect differential effects of the subcutaneous microenvironment. In general, tumors formed by cells expressing the single and double tyrosine mutants were much larger than those generated by cells transfected with wild type CEACAM1-4S or empty vector, a trend consistent with the greater rate of proliferation shown in vitro by tyrosine mutants. When viewed as a whole, these findings show that the phenotypes manifested in vitro by cells expressing transmembrane domain mutants are recapitulated in vivo, suggesting these changes are intrinsic to the transmembrane domain and are not the result of extrinsic factors e.g., microenvironment, that change the phenotypic effects of CEA-CAM1-4S by altering its proteomic context. The decrease in soft agar growth produced by glycine mutations and the increase by tyrosine mutations relative to the wild type CEACAM1-4S would classify these, respectively, as loss-of-function and gain-of-function mutations [47, 48, 49] . Gain-of-function mutations are usually dominant, a characteristic apparent for the glycine/tyrosine mutants of CEACAM1-4S where the growth enhanced phenotype of tyrosine mutants was dominant over the growth inhibited phenotype resulting from glycine mutations. Based on the known functions of GXXXG motifs, it seems likely that the loss-of-function caused by glycine mutations is related to changes in helix-helix interactions that destabilize rather than prevent the formation of cis-dimers [28, 50, 51] , a possibility consistent with the fact that single G424L or G432L mutations failed to disrupt dimerization ( Figure 10 ) but did cause a decrease in anchorage independent growth. Also noteworthy is the marked decrease in cisdimerization ( Figure 10 ) when mutations were introduced at both G424 and G432, a result that suggested G424-AIA-G428 was a functional GXXXG motif capable of directing cis-dimerization in the absence of the two motifs centered on G432 and necessary to sustain the level of growth produced by wild type CEACAM1-4S. Although this conclusion is seemingly at odds with the 9 prediction algorithms that placed G424 outside the transmembrane domain (Table 1) , we suggest that a combination of structural and functional data provides a more accurate location of the transmembrane domain, one that incorporates the GAIAG motif at the N-terminus. Similar reasoning can be applied to the C-terminus where 5 of 12 transmembrane prediction algorithms placed Y448 outside the transmembrane domain. However, mutation of either Y445 or Y448 led to the soft agar growth enhanced phenotype, indicating first, that Y448 was functionally equivalent to Y445 and thus likely to be located within the transmembrane domain and second, that both tyrosines were required to sustain anchorage independent growth at the level produced by wild type CEACAM1-4S. However, whether Y448 was or was not in the transmembrane was moot since single Y445V or Y448V mutations produced the growth-enhanced phenotype in soft agar. Put another way, the presence of both tyrosines appeared to be necessary to suppress anchorage independent growth, a suppressive effect that apparently involved more than interactions between aromatic side chains since in both single tyrosine to valine mutants (VFLY or YFLV) the remaining tyrosine was paired with F446. Whether the two tyrosines without phenylalanine would be able maintain the wild type CEACAM1-4S phenotype is an open question that is beyond the scope of the present investigation. While the mechanism behind the gain-of-function produced by the Y mutations is less clear, there is increasing evidence supporting the functional importance of interactions between transmembrane aromatic amino acids [41] . Although aromatic residues make up only a small percentage of the amino acids in any given protein, they are generally the most highly conserved residues. Interactions between the aromatic rings of phenylalanine, tryptophan (W) or tyrosine are thought to involve pi-stacking, a process that creates an attractive force when aromatic rings assume energetically favored stacking geometries. Accumulating evidence suggests that pi-stacking plays an important role in molecular recognition and self assembly either by contributing energy for driving self assembly or by providing directionality and orientation through stacking geometries [52, 53] . When a bacterial transmembrane database was statistically analyzed by Sal-Man et al, these investigators found that aromatic pairs (WXXW or YXXY) were significantly over-represented compared to their predicted frequency, suggesting a functional role for these sequences [52] . The stabilization of transmembrane domain selfassociation following substitution of tyrosine for glutamine (Q) and serine (S) in the known dimerization motif, QXXS, provided support for this idea and led these authors to conclude that stabilization of transmembrane associations by aromatic residues may be a general mechanism for generating specificity in transmembrane-transmembrane interactions. In considering which amino acid to substitute for the transmembrane tyrosine residues, phenylalanine would intuitively seem to be the logical When separated on native gels, wild type CEACAM1-4S and the single G mutants migrated with an apparent molecular mass that was approximately 100 kDa higher than the double glycine mutant. doi:10.1371/journal.pone.0029606.g010 choice. However, based on recent reports, tyrosine to phenylalanine mutations may or may not disrupt function. Stevens et al [54] reported that a YS/FA mutation of the IgM transmembrane domain had no effect on anti-Ig induced signaling as measured by the activation of tyrosine phosphorylation and did not disrupt the association between IgM and its signaling partners, Ig-alpha/Igbeta. In contrast, a YS/VV mutation diminished both signaling and association with Ig-alpha/Ig-beta. In cases where phenylalanine substitutions fail to mimic tyrosine, the lack of a properly spaced hydroxy group may be the reason [41] . Based on these considerations, it was decided to introduce tyrosine to valine mutations. GXXXG motifs and aromatic amino acids are common features of the transmembrane domains of many different plasma membrane proteins with GXXXG motifs most often located at the N-terminus [55, 56, 57] and Y/F/W aromatic amino acids at the C-terminus of the transmembrane domain [58, 59] . Many of the proteins also have Sternberg-Gullick dimerization motifs, a family of sequences discovered in the transmembrane domain of tyrosine kinase growth factor receptors [60] . Members of this family, which includes a subset of GXXXG/AXXXG/SXXXG dimerization motifs, are composed of 5 amino acids arranged with G, A, S, T or P in the first position (N-terminus), an A, V, L or I in the fourth position and G or A in the fifth position. Our findings demonstrate that a single amino acid substitution in the transmembrane domain of CEACAM1-4S can produce dramatic effects on cell proliferation, anchorage independent growth and in vivo tumorigenicity. It seems clear that the transmembrane domain and more specifically GXXXG motifs and tyrosine residues make a significant contribution to the functionality of CEACAM1-4S and by extension, to other transmembrane proteins with similar characteristics. Further studies are needed to define the downstream signaling events impacted by re-expression of CEACAM1-4S or by transmembrane domain mutations, studies that should provide valuable insights into events controlled by transmembrane domain mediated interactions. Of particular interest will be the identity of pathways inactivated/activated by tyrosine mutations that lead to the positive growth effects of tyrosine mutations and the ability of these pathways to counteract the tumor suppressor activity of CEACAM1-4L, the larger splice variant with the same transmembrane domain as the 4S isoform.

Human Subtilase SKI-1/S1P Is a Master Regulator of the HCV Lifecycle and a Potential Host Cell Target for Developing Indirect-Acting Antiviral Agents

HCV infection is a major risk factor for liver cancer and liver transplantation worldwide. Overstimulation of host lipid metabolism in the liver by HCV-encoded proteins during viral infection creates a favorable environment for virus propagation and pathogenesis. In this study, we hypothesize that targeting cellular enzymes acting as master regulators of lipid homeostasis could represent a powerful approach to developing a novel class of broad-spectrum antivirals against infection associated with human Flaviviridae viruses such as hepatitis C virus (HCV), whose assembly and pathogenesis depend on interaction with lipid droplets (LDs). One such master regulator of cholesterol metabolic pathways is the host subtilisin/kexin-isozyme-1 (SKI-1) – or site-1 protease (S1P). SKI-1/S1P plays a critical role in the proteolytic activation of sterol regulatory element binding proteins (SREBPs), which control expression of the key enzymes of cholesterol and fatty-acid biosynthesis. Here we report the development of a SKI-1/S1P-specific protein-based inhibitor and its application to blocking the SREBP signaling cascade. We demonstrate that SKI-1/S1P inhibition effectively blocks HCV from establishing infection in hepatoma cells. The inhibitory mechanism is associated with a dramatic reduction in the abundance of neutral lipids, LDs, and the LD marker: adipose differentiation-related protein (ADRP)/perilipin 2. Reduction of LD formation inhibits virus assembly from infected cells. Importantly, we confirm that SKI-1/S1P is a key host factor for HCV infection by using a specific active, site-directed, small-molecule inhibitor of SKI-1/S1P: PF-429242. Our studies identify SKI-1/S1P as both a novel regulator of the HCV lifecycle and as a potential host-directed therapeutic target against HCV infection and liver steatosis. With identification of an increasing number of human viruses that use host LDs for infection, our results suggest that SKI-1/S1P inhibitors may allow development of novel broad-spectrum biopharmaceuticals that could lead to novel indirect-acting antiviral options with the current standard of care.

Hijacking of host lipids and their biosynthetic pathways is a common strategy for microbial infection. Human enveloped viruses including hepatitis C virus (HCV) and human immunodeficiency virus (HIV)-1 use cholesterol-rich lipid rafts for entry [1, 2] , assembly [3] , and/or replication [2, 4] . Lipid droplets (LDs), once considered static storage vesicles for host lipids, are now appreciated as dynamic organelles [5] that are also utilized in the lifecycles of pathogenic human viruses including rotavirus (RV) [6] , dengue virus (DV) [7] , and HCV [8] . HCV in particular requires host LDs for assembly of nascent viral particles [9] [10] [11] . HCV is a globally important human pathogen afflicting more than 170 million people worldwide [12, 13] . HCV, a hepacivirus member of the Flaviviridae family and an enveloped virus, is encoded by a single-stranded positive-sense RNA genome [14] . Viral RNA is directly translated by the host machinery into a single polyprotein, which is cleaved by host and virus-encoded proteases to release the individual structural (core, E1, and E2) and non-structural (NS) proteins (p7, NS2, NS3, NS4A, NS4B, NS5A, and NS5B) [15] . During infection, HCV-encoded proteins promote reorganization and accumulation of LDs in the perinuclear region of the cell [16] . The HCV core protein is targeted to LDs [17] and orchestrates the assembly and release of infectious viral particles during the late stages of infection [18] . Hence, disrupting the interaction of the HCV core protein with LDs compromises this essential stage within the HCV lifecycle [8, 10, 11] . Several host metabolic pathways tightly control cellular lipid synthesis. Targeted disruption of these pathways [19] [20] [21] by HCVencoded proteins has been linked with liver steatosis [22, 23] in HCV-infected individuals. Importantly, there is a correlation between the degree of steatosis and both the severity of chronic HCV infection [13, 24] and the response to treatment with pegylated-interferon-a and ribavirin [25, 26] . Overstimulation of host lipid metabolism by HCV during infection is achieved through a variety of molecular mechanisms (reviewed in [27] , [28] , and [29] ). For example, HCV employs multiple strategies to activate the sterol regulatory element binding protein (SREBP) pathway, which is important for regulation of host lipid homeostasis [19, [30] [31] [32] . SREBPs are endoplasmic reticulum (ER), membrane-anchored transcription factors that respond to changes in intracellular sterol levels through interactions with sterol-sensing proteins (reviewed in [33] ). When sterol levels are high, SREBPs are retained as inactive precursors in the ER [34, 35] . Under low sterol conditions, SREBPs are escorted to the Golgi, where two resident endoproteases (subtilisin kexin isozyme/site-1 protease (SKI-1/S1P) and SREBP Site-2 protease (S2P); reviewed in [36] and see below) cleave the precursor polypeptide SREBP, allowing the release of transcriptionally active SREBP molecules from the ER [37, 38] . The released SREBP fragment migrates to the nucleus and binds to sterol response elements in the promoters of cholesterol and fatty acid biosynthetic target genes, and it activates their transcription [39] [40] [41] . Activation of the SREBP pathway by HCV aids the virus lifecycle and may ultimately promote the development of steatosis and liver disease in chronically infected individuals [42, 43] . Human site-1 protease (S1P, MEROPS S08.8063), also widely known as subtilisin/kexin-isozyme-1 (SKI-1), is a membranebound subtilisin-related serine endoprotease that belongs to a group of nine mammalian proprotein convertases (PCs) in family S08 [44] [45] [46] . SKI-1/S1P displays unique substrate specificity among the PC members by showing preferred cleavage after non-basic amino acids [47] . SKI-1/S1P cleaves at the carboxyterminus of the peptidyl sequence Arg/Lys-Xaa 3 -Xaa 2 -Leu/ Ser/Thr [48] , where Xaa 3 is any amino acid except Cys at the P3 position of the scissile peptide bond and Xaa 2 is a hydrophobic amino acid containing an alkyl side chain at the P2 position [49] . In addition to the proteolytic processing of transcription factors [36] , SKI-1/S1P participates in the proteolytic activation of viral-envelope glycoproteins of the Lassa virus [50] , the lymphocytic choriomeningitis virus [51] , and the Crimean Congo hemorrhagic fever virus [52] . Importantly, endoproteolytic processing of these viral glycoproteins by SKI-1/S1P is a critical step for the production of infectious progeny viruses, suggesting that SKI-1/S1P may represent an attractive target for therapeutic intervention against human pathogenic arenaviruses [47, 48, [50] [51] [52] . Given that SKI-1/S1P-dependent proteolytic cleavage of SREBPs is a master molecular switch for controlling host cell cholesterol homeostasis [41, 53] , we hypothesized that inhibiting SKI-1/S1P endoproteolytic activity in the secretory pathway would prevent HCV hijacking of host lipid metabolic pathways and thus compromise the virus lifecycle. Because of our previous success with engineering serine protease inhibitors (serpins) to develop effective and selective PC inhibitors [54, 55] , we hypothesized that engineering a naturally occurring serpin scaffold could also provide a powerful approach for developing selective SKI-1/S1P inhibitors. Serpins differ from non-serpin inhibitors in that they require a large inhibitor conformational change in order to trap proteases in an irreversible complex [56] . The conformational change is initiated by reaction of the active serine of the protease with the reactive center loop (RCL) of the serpin, which results in a covalent species involving an acyl ester linkage to the cO of the protease active site serine [57] . This cleaves the RCL, which then moves 71 Å to the opposite pole of the serpin, taking the tethered protease with it [57] . We selected a Drosophila serpin, Spn4A, as a prototype macromolecular inhibitor scaffold. The identification of Spn4A by our group as the most potent natural serpin inhibitor of the human PC furin (K i ,13 pM [58] ) provides us with a novel and unique molecular tool for dissecting the contribution of SKI-1/ S1P-dependent proteolytic activity in the secretory pathway to viral infection of eukaryotic cells. We report that engineering of the RCL of Spn4A to mimic the consensus sequence Arg/Lys P4 -Xaa 3 -Xaa 2 -Leu/Ser/Thr P1Q for cleavage by SKI-1/S1P (Spn4A RCL: Arg P4 -Arg-Lys-Arg P1Q -. Arg P4 -Arg-Leu-Leu P1Q ) resulted in the development of Spn4A-RRLL, a novel, selective, and effective serpin-based inhibitor of SKI-1/S1P. We demonstrated the antiproteolytic and anti-HCV activities of our new recombinant adenovirus-expressing Spn4A-RRLL ''secreted'' (s) variant directed at the secretory pathway SKI-1/S1P. Expression of Spn4A.RRLL(s) in Huh-7.5.1 cells results in a strong inhibition of the SKI-1/S1P-mediated activation of SREBP-1 and downregulation of SREBP target gene products. As hypothesized, inhibiting SKI-1/S1P activity robustly blocked HCV infection of Huh-7.5.1 cells in a dose-dependent manner. We found that specific inhibition of SKI-1/S1P activity by Spn4A.RRLL(s) dramatically reduced the abundance of LDs in hepatoma cells. Use of the specific active site-directed small-molecule inhibitor of SKI-1/S1P, PF-429242, confirmed the results of our studies with the protein-based therapeutic Spn4A.RRLL(s), with a robust inhibition of HCV infection. The results of our studies contribute to our understanding of the HCV lifecycle and HCV-associated steatogenesis and to efforts in developing novel host-directed antiviral therapeutic agents against HCV. In addition, with the finding that an increasing number of human enveloped viruses employ host LDs for infection [6, 7] , our results suggest that SKI-1/S1P-directed inhibitors may allow the development of novel broad-spectrum antiviral agents. Protein engineering of the Spn4A scaffold and drug delivery strategy to target secretory pathway SKI-1/S1P Our previous studies have demonstrated that Spn4A architecture can inhibit two evolutionary divergent members of the PC family (furin and PC2) [58] . We selected this novel Drosophila melanogaster serpin scaffold to engineer a novel protein-based inhibitor directed at the PC SKI-1/S1P. First, we cloned our pre-His/FLAG-tagged Chronic hepatitis C virus (HCV) infection is one of the leading causes of liver cancer and liver transplantation worldwide. No vaccine is available for preventing the spread of HCV, and the current therapeutic regimen is only moderately effective and causes serious side effects. New antiviral agents are required to treat HCV infection, but the high mutation rate of HCV hinders the effectiveness of virus-specific inhibitors. Targeting the host enzymes required for HCV to replicate offers a promising new direction for antiviral therapy. During infection, HCV promotes excessive fat accumulation in the liver, which benefits the virus as this promotes formation of lipid droplets, a cellular organelle essential for assembly of new HCV infectious viral particles. Here, we report the development of a specific inhibitor targeting SKI-1/S1P, a host enzyme required for lipid production in human cells. We show that inhibiting SKI-1/S1P activity in human liver cells effectively blocks lipid droplet formation and HCV infection. Many prevalent human viruses, such as dengue, rotavirus, and hepatitis B virus, hijack host lipid metabolic pathways similar to those targeted by HCV to complete their lifecycle. Thus, we propose that cellular SKI-1/S1P is a potential target for developing desperately needed novel broad-spectrum antiviral drugs. Spn4A construct [58] into an adenoviral shuttle vector to generate Spn4A.RRKR(r) ( Figure 1A ). Spn4A is a unique ''retained'' (r) serpin that presents, at its C-terminus, an HDEL sequence ( Figure 1A) , a functional variant of the C-terminal KDEL retention signal that directs secretory protein retention in the ER [59] . Because SKI-1/S1P cleavage of SREBP substrates occurs in the Golgi apparatus [35] , we next needed to generate a ''secreted'' (s) variant of Spn4A, Spn4A.RRKR(s) ( Figure 1A ). We hypothesized that only an Spn4A (s) variant, which traffics through the late secretory pathway prior to secretion in the extracellular space, would encounter active SKI-1/S1P molecules in the early Golgi compartment. This was accomplished by inserting a stop codon before the C-terminal ER-retention signal, HDEL. Next, we employed site-directed mutagenesis to optimize the interactions between the RCL of Spn4A and the substrate binding sites of SKI-1/S1P. Residues at positions P2 and P1 of the Spn4A RCL ''bait'' region Arg P4 -Arg P3 -Lys P2 -Arg P1Q were substituted to generate Arg P4 -Arg P3 -Leu P2 -Leu P1Q , thereby mimicking the Lassa virus glycoprotein precursor GP-C cleavage site [50] ( Figure 1A and 1B: Spn4A.RRLL(r) and Spn4A.RRLL(s), respectively). To test the serpin-like properties and antiviral activities of our Spn4A variants in cellulo, recombinant adenoviruses (Ad) expressing the Spn4A constructs Ad-Spn4A.RRLL(r) and Ad-Spn4A.RRLL(s) were produced as described previously [55] . As adenoviruses display strong tropism for the liver [60] , the major site of HCV infection [13] , these recombinant adenoviruses are especially useful molecular tools for HCV research, including the use of human hepatoma Huh-7.5.1 cells, which support robust HCV infection in cellulo [61] . Robust intracellular expression and differential secretion of adenovirus-expressed Spn4A.RRLL(r) and Spn4A.RRLL(s) in human hepatoma Huh-7.5.1 cells The level of expression of Spn4A variants was first optimized by infecting human hepatoma Huh-7.5.1 cells (highly permissive for HCV JFH-1 infection) with different multiplicity of infection (moi) of recombinant adenovirus expressing intracellularly retained Spn4A.RRLL(r). Cellomics HCS analysis revealed that over 90% of these Huh-7.5.1 cells expressed Spn4A.RRLL(r) at a moi of 50 ( Figure S1 ). Cellomics HCS was also used to determine if cell death occurs following 2 days of pre-treatment with Spn4A.RRLL(r) or Spn4A.RRLL(s) compared to the control (Ad-Empty) followed by 72 hours of HCV infection as employed in the experiments below ( Figure S2 ). We observed no significant reductions in total cell numbers under these experimental conditions. The high-content analysis of cell death was confirmed using an MTS-based cell viability assay (data not shown). The lack of detectable toxicity induced by Ad-Spn4A.RRLL(r) and Ad-Spn4A.RRLL(s) up to a moi of 50 showed that these variants can be tested over a very wide dynamic range in hepatoma cells. The serpin-secretion profile in Huh-7.5.1 cells was examined by Western blotting of cell lysates and extracellular media. As expected, a prominent 45-kDa band was detected in Ad-Spn4A.RRLL(r)-infected and Ad-Spn4A.RRLL(s)-infected cell lysates using anti-FLAG antibody (Figure 2A, lanes 2 and 3) . Furthermore, as hypothesized, the 45-kDa band of Spn4A.RRLL(r) was found only in lysed cell extracts and was not found secreted into extracellular media (Figure 2A Serpin-like properties and intrinsic specificity of Spn4A.RRLL(r) and Spn4A.RRLL(s) A stable, acyl-enzyme complex is formed between a protease and a functional inhibitory serpin following RCL cleavage. This allows for detection of the high molecular weight, heat-stable, and SDS-stable enzyme-inhibitor (EI) complex by standard SDS- Spn4A.RRKR(r) encodes for the naturally occurring serpin Spn4A, isolated from Drosophila melanogaster, with potent inhibitory activity against the human proprotein convertase furin. Spn4A.RRKR(r) contains the alpha-1 antitrypsin signal peptide (SP) at the N-terminus followed by a tandem His-tag (HHHHHH) and FLAG-tag (DYKDDDDK) sequence (HF). The P4 -P1 furin cleavage sequence in the RCL is Arg-Arg-Lys-Arg. Spn4A.RRKR(r) also contains the His-Asp-Glu-Leu (HDEL) ER retention motif (r) at the C-terminus. The secreted (s) serpin, Spn4A.RRKR(s), contains a stop codon before the HDEL signal. The RCL of Spn4A-RRKR(r) and (s) was modified to mimic the predicted SKI-1/S1P target cleavage site present in the Lassa virus glycoprotein pre-GP-C, which is Arg-Arg-Leu-Leu. Thus, Spn4A.RRLL(r), which is also retained in the ER, encodes the P4 -P1 Arg-Arg-Leu-Leu cleavage recognition sequence in the RCL. Spn4A.RRLL(s) contains a stop codon before HDEL, allowing the serpin to be secreted. (B) In silico homology model of the Spn4A.RRLL(r) variant was generated as described in the Materials and Methods. Ribbon diagram of the molecular model was generated using Pymol. The side chains of the RRLL residues within the flexible ''bait region'' of the RCL are shown as sticks in wheat colour. Sheet A is shown in yellow, sheet B is in blue, and sheet C is in cyan. Alpha-helices are red and loops are green. doi:10.1371/journal.ppat.1002468.g001 PAGE and Western blot [54, 55, 58, 62] . To determine if Spn4A.RRLL(s) is a functional and selective inhibitor of SKI-1/ S1P, recombinant Spn4A variants (furin-and SKI-1/S1P-directed inhibitors) were expressed in Huh-7.5.1 cells for 72 hours. Cell media and extracts containing recombinant serpins were then harvested and incubated with purified recombinant His-tagged human SKI-1/S1P or furin ( Figure 2B ) as described previously [47, 54, 58] . Reaction mixtures were analyzed by Western blot and probed for EI complex formation with anti-His antibody ( Figure 2B ) and anti-FLAG antibody ( Figure S3 ). The results shown in Figure 2B clearly demonstrate EI complex formation between recombinant SKI-1/S1P and Spn4A.RRLL(s) in cell media and in cell extracts (lane 5, upper and lower panels). As expected, Spn4A.RRLL(s) did not form a complex with furin ( Figure 2B , lane 10, upper and lower panels), whereas the furindirected serpin Spn4A.RRKR(s) formed an EI complex with furin but not with SKI-1/S1P in cell media and extracts ( Figure 2B , lanes 9 and 7, respectively, upper and lower panels). Lysed cellular extracts expressing Spn4A.RRLL(r) also demonstrated EI complex formation with SKI-1/S1P ( Figure 2B , lane 12, bottom panel). The results of our biochemical analysis confirmed the serpin-like properties of recombinant Spn4A.RRLL(r) and (s) biosynthesized in human hepatoma cells and the selectivity of Spn4A.RRLL(s) against SKI-1/S1P. Importantly, Spn4A.RRLL(s) inhibits SKI-1/ S1P by a suicide substrate mechanism and forms a kinetically trapped heat-and SDS-stable complex with SKI-1/S1P as is characteristic of other physiological serpin-protease pairs [58, 63] . Spn4A.RRLL(s) is a potent inhibitor of SKI-1/S1P-mediated endoproteolytic cleavage of SREBPs, of their downstream effector gene expression, and of intracellular cholesterol-ester accumulation To confirm that expression of Spn4A.RRLL(s) in Huh-7.5.1 cells inhibits endogenous SKI-1/S1P-mediated cleavage of SREBP molecules, we examined nuclear SREBP-1 protein levels in cells infected with adenovirus-expressed Spn4A variants. As a positive control, we also treated cells with the selective, reversible, and competitive small-molecule inhibitor of SKI-1/S1P: PF-429242 [64, 65] . This compound was recently synthesized and characterized both in vitro and in vivo for its anti-lipidemic properties including efficient inhibition of nuclear SREBP accumulation. As previously described [64, 66] , the calpain inhibitor, alpha-N-acetyl-Leu-Leu-Nle-CHO (ALLN), was employed to facilitate the detection and accumulation of the Nterminal fragment of SREBP-1 in the nucleus. An anti-fibrillarin antibody was used to positively identify the nuclear fractions ( Figure 3A ) [67] . As expected, Western blotting of nuclear extracts from cells treated with 10 mM of PF-429242 (PF-429242 + ALLN), using an antibody against the N-terminal fragment of SREBP-1, revealed a complete block of SREBP-1 accumulation in the nucleus ( Figure 3A ). Nuclear extracts from cells infected with Ad-Spn4A.RRLL(s) (RRLL(s) + ALLN) also exhibited a dramatic decrease in nuclear SREBP-1 accumulation when compared to Ad-Empty (control + ALLN) and Ad-Spn4A.RRLL(r) (RRLL(r) + ALLN)-infected cells. These results confirm that expression of recombinant Spn4A.RRLL(s) in the secretory pathway of Huh-7.5.1 cells inhibits SREBP-1 processing by SKI-1/S1P. Next, to determine whether the Spn4A.RRLL(s)-mediated reduction in nuclear SREBPs was associated with a concomitant decrease in the protein levels in SREBP-target genes, we examined the fate of three SREBP-dependent gene products, SREBP-2, LDLR, and PCSK9. We investigated these host cell proteins because of their proposed contribution to HCV entry (LDLR and PCSK9) and propagation (SREBP-2) [19, 68, 69] . A time course analysis of cells expressing Spn4A.RRLL(s) in complete media showed the most significant block in SREBP-regulated LDLR expression after 72 hours ( Figure S4 ). As HCV is known to induce SREBP activation [19, 31, 32] , we then analyzed the expression of SREBP-regulated proteins under SREBP-activated conditions ( Figure 3B ). Huh-7.5.1 cells were depleted of exogenous sterols for 24 hours to induce SREBP transport from ER-to-Golgi prior to infection with Ad-Empty, Ad-Spn4A.RRLL(r), or Ad-Spn4A. RRLL(s). The levels of LDLR, PCSK9, and SREBP-2 (all regulated by nuclear SREBPs [70] [71] [72] [73] ) were then measured using Western blot analysis of lysed cell extracts ( Figure 3B ). After 72 hours of Spn4A.RRLL(s) expression, mature LDLR (160 kDa) levels were reduced by 74% compared to Ad-Empty-treated cells. Similarly, an 85% block in mature PCSK9 (60 kDa) expression and a 79% reduction in full-length SREBP-2 expression were observed. No significant reductions in LDLR or SREBP-2 levels following Spn4A.RRLL(r) treatment were observed. Interestingly, a significant reduction in PCSK9 expression was detected in Spn4A.RRLL(r)-expressing cells ( Figure 3B ). The expression of btubulin and of the Golgi marker GM130 were not affected by Spn4A.RRLL(r) or Spn4A.RRLL(s) expression ( Figure 3B ). These results confirm that expression of Spn4A.RRLL(s) in Huh-7.5.1 cells specifically inhibits the SREBP pathway including target genes identified as cellular cofactors affecting HCV infection. A critical function of the SREBP pathway and the genes that it regulates is to control lipid homeostasis [36] . We investigated the impact of inhibiting SKI-1/S1P using both Spn4A.RRLL(s) and PF-429242 on total intracellular lipid levels, specifically cholesterol, cholesterol-esters, triglycerides, and phospholipids ( Figure 3C and 3D). Among the cell lipids examined, Spn4A.RRLL(s) and PF-429242 had the most dramatic impact on cholesterol-ester levels, a major constituent of cellular LDs [5] ; these were reduced by 74% in Spn4A.RRLL(s)-treated cells compared to control (Ad-Empty)-treated cells ( Figure 3C ). Similarly, PF-429242 reduced cholesterol-ester levels by , 63% compared to control cells treated with DMSO ( Figure 3D ). A 14% reduction in triglycerides was also induced by Spn4A.RRLL(s), although this reduction did not reach significance, whereas PF-429242 caused a significant 51% reduction in total intracellular triglycerides. A significant 5% reduction in free cholesterol levels was also observed in Spn4A.RRLL(s)-treated cells and a 25% reduction was observed in PF-429242-treated cells compared to respective controls. No significant reductions in phospholipid levels were detected ( Figure 3C and 3D). These results suggest that sustained inhibition of SKI-1/S1P-mediated cleavage and activation of SREBP causes increased cellular utilization of lipid stores. Expression of secretory pathway Spn4A.RRLL(s) in Huh-7.5.1 cells dramatically reduces the abundance of cellular lipid storage droplets LDs are dynamic intracellular lipid storage compartments made up of triglyceride and cholesterol esters, surrounded by a phospholipid membrane and associated with specific marker proteins including adipose differentiation-related protein (ADRP), also known as perlipin 2 [74] . Because SREBP activation controls the expression of genes directly involved in intracellular fatty acid and cholesterol biosynthesis (reviewed in [36] ) and because cholesterol-ester levels were reduced by Spn4A.RRLL(s), we investigated the effect of serpin-mediated SKI-1/S1P inhibition on cellular LD abundance. Fluorescence microscopy was used to determine the relative abundance of LDs stained with BODIPY 493/503 in Huh-7.5.1 cells infected with Ad-Empty (control), Ad-Spn4A.RRLL(r), and Ad-Spn4A.RRLL(s). After 72 hours, the level of BODIPY-stained LDs in Spn4A.RRLL(s)-expressing cells was dramatically reduced compared to empty vector-treated cells ( Figure 4A ). By contrast, Spn4A.RRLL(r)-expressing cells had no apparent reduction in LD size or abundance compared to controltreated cells ( Figure 4A ). Quantification of confocal microscopy images demonstrated that, on average, LD abundance was reduced by 80% in Spn4A.RRLL(s)-expressing cells compared to controls ( Figure 4B ). The effect of Spn4A.RRLL(s) expression on cytosolic LD abundance was confirmed by visualizing the LD marker ADRP/perilipin 2 using confocal microscopy ( Figure 4C ) and by using quantitative Western blot ( Figure 4D ). Spn4A.RRLL(s) was observed to reduce ADRP/perilipin 2 protein expression by 50% compared with control cells ( Figure 4D ), whereas there was no reduction in ADRP/perlipin 2 expression following Spn4A.RRLL(r) treatment. We subsequently confirmed that reduced cellular LD levels were due to inhibition of SKI-1/S1P using 10 mM PF-429242 whereupon ADRP/perilipin 2 levels were reduced by 62% compared to control DMSO-treated cells ( Figure 4E ). These results confirm that Spn4A.RRLL(s)-and PF-429242-mediated inhibition of SKI-1/S1P enzymatic activity dramatically reduces intracellular LD abundance in Huh-7.5.1 cells. Expression of secretory pathway Spn4A.RRLL(s) in Huh-7. 5 Since the SREBP signaling pathway is induced by HCVencoded proteins during infection [19, [30] [31] [32] , we tested the effect of inhibiting this pathway on the HCV lifecycle in human hepatoma cells. Huh-7.5.1 cells were treated with increasing moi (1-50) of Ad-Spn4A.RRLL(r), Ad-Spn4A.RRLL(s), or Ad-Empty (control) for 48 hours in complete media with or without exogenously added sterols followed by 72 hours of infection with HCV. The number of HCV-infected cells, as evidenced by positive core protein expression, was measured using Cellomics HCS ( Figure 5A ). It was determined that Spn4A.RRLL(s) expression inhibited HCV infection in a dose-dependent manner compared to control-treated cells. HCV infection was not significantly reduced in cells infected with Ad-Spn4A.RRLL(s) at a moi of 1. A moi of 12.5, however, caused a 40% reduction, a moi of 25 caused a 60% reduction, and a moi of 50 caused a 75% reduction in the number of HCV-infected cells compared to controls. Spn4A.RRLL(r) had no significant impact on HCV infection up to adenovirus moi 50 when compared to the control ( Figure 5A ). Supplementing sterol and lipid metabolites significantly restored infectivity when cells were treated with moi 50 of Ad-Spn4A.RRLL(s), where a 2-fold increase in HCV infection compared to non-supplemented cells was observed ( Figure 5A ). These results show that the anti-HCV activity of Spn4A.RRLL(s) is, at least in part, associated with its capacity to decrease intracellular lipid stores within the host cells. The anti-HCV properties of Spn4A.RRLL(s) were confirmed and extended by further virological studies on cells pre-treated with our serpin-based inhibitors for 48 hours prior to 72 hours of HCV infection ( Figure 5B and 5C). First, quantitative Western blot analysis revealed a 64% reduction in the expression of intracellular HCV core protein in Spn4A.RRLL(s)-expressing cells compared to the control ( Figure 5B ). Similarly, Spn4A.RRLL(s) treatment was found to reduce extracellular infectious HCV titers by 76% ( Figure 5B) . A 65% reduction in intracellular HCV RNA levels was also observed using quantitative PCR (QPCR analysis ( Figure 5C ). Spn4A.RRLL(r) expression did not significantly impact any aspect of the HCV lifecycle examined ( Figure 5A , 5B and 5C). These results demonstrate that inhibition of SKI-1/ S1P-mediated proteolytic activation of SREBP molecules using secretory pathway protein-based inhibitors is an effective antiviral strategy to robustly block HCV infection in Huh-7.5.1 cells. To determine if a decrease in intracellular HCV RNA in Spn4A.RRLL(s)-treated cells ( Figure 5C ) is due to reduced viral replication or alternatively due to reduced HCV entry, we examined cells transfected directly with total genomic HCV RNA for 3 days following 48 hours of adenovirus-mediated serpin expression. We found that under these experimental conditions, when receptor-mediated HCV entry was bypassed, no significant . Acquisition and analysis were performed using the same intensity and threshold settings across all images. (C) The LD marker ADRP was detected in cells treated with Spn4A.RRLL(r), Spn4A.RRLL(s), and Ad-Empty (control) using rabbit anti-ADRP antibody (green), and images were obtained using an Olympus Fluoview FV1000 laser scanning confocal changes in intracellular HCV RNA levels were detected by QPCR ( Figure 5C ). Examination of total cell extracts by Western blot revealed that HCV core levels are not reduced following HCV RNA transfection in serpin-treated cells ( Figure S5) , confirming that Spn4A.RRLL(s) does not interfere with HCV replication when HCV entry is bypassed. The impact of Spn4A.RRLL(s) treatment on HCV replication was further investigated using HCV subgenomic replicons [75] . Human hepatoma cells harbouring stable HCV replicons encoding wild-type NS5A (Huh.8 cells) or NS5A with an adaptive mutation (Huh.2 cells) were treated with recombinant adenoviruses for 5 days. Total RNA levels were then harvested and the level of HCV RNA was quantified using QPCR analysis. No significant differences were observed between HCV replicon levels treated with Spn4A.RRLL(s), Spn4A.RRLL(r), and the control (Ad-Empty) ( Figure 5D ). These results confirm that Spn4A. RRLL(s) does not inhibit HCV replication. These findings, in addition to the measured decrease in LDLR expression presented in Figure 3B , strongly suggest that the robust reduction in intracellular HCV RNA levels observed in Spn4A.RRLL(s) pretreated cells prior to HCV infection is, at least in part, due to reduced viral attachment or entry. Next, we wanted to test whether extracellularly applied PF-429242 would effectively inhibit the endoproteolytic activity of secretory pathway SKI-1/S1P and reduce HCV infection in Huh-7.5.1 cells. First, using an MTS-based cell viability assay, we confirmed that no major cytotoxic effects occur in Huh-7.5.1 cells treated with up to 50 mM of PF-429242 ( Figure S6 ). Next, cells were treated with increasing concentrations (0.05 mM to 50 mM) of PF-429242 for 24 hours before the cell media was replaced and cells were infected for 48 hours with HCV. The number of HCVinfected cells, indicated by positive core protein expression, was measured using Cellomics HCS ( Figure 6A Because PF-429242 and Spn4A.RRLL(s) decrease the abundance of LD components, we hypothesized that HCV assembly, rather than HCV replication, can also blocked by SKI-1/S1P inhibition. To support this hypothesis, Huh.8 and Huh.2 cells were treated with 10 mM PF-429242 for 72 hours. No significant changes in HCV subgenomic RNA levels in either cell line ( Figure 6C ) treated with PF-429242 were observed (compared to DMSO-treated control). In summary, with the lack of effect of PF-429242 on HCV replicon levels and comparing the two sets of data presented in Figure 6B , which demonstrate a strong antiviral effect of PF-429242 when added either pre-or post-HCV inoculation, we can propose that pharmacological inhibition of SKI-1/S1P endoproteolytic activity by PF-429242 impacts late assembly stages of the HCV lifecycle. It is now well established that hijacking of host-cell biosynthetic pathways by human enveloped viruses is a shared molecular event essential for the viral lifecycle [76] . The next frontier is identifying common, critical, host cell pathways that are hijacked by pathogenic human viruses, in order to develop broad-spectrum, host-directed antivirals with novel mechanisms of action. In this study, we hypothesized that targeting cellular enzymes acting as master regulators of lipid homeostasis could represent a powerful approach to developing a novel class of antiviral agents against infection associated with human enveloped viruses such as HCV, whose replication and pathogenesis depend on the interaction with lipid droplets (LDs) [8] . In the case of HCV, overstimulation of host lipid metabolism in the liver during viral infection promotes cholesterol intracellular storage in host LDs, a critical cellular event for the HCV lifecycle that leads to steatosis of the liver in HCV-infected patients [8, 18, 77] . One such master regulator of cholesterol metabolic pathways is the host proprotein convertase SKI-1/S1P [36, 78] . SKI-1/S1P plays a critical role in the proteolytic activation of SREBPs, which control expression of key enzymes of cholesterol and fatty-acid biosynthesis [41, 53] . Here, we report that strategic manipulation of cellular SKI-1/S1P activity levels by protein-based or small-molecule protease inhibitors provides a means of effectively microscope. (D) Huh-7.5.1 cells infected with Ad-Empty (control), Ad-Spn4A.RRLL(r), or Ad-Spn4A.RRLL(s) for 72 hours were harvested and subjected to SDS-PAGE and Western blot analysis. Mouse anti-ADRP antibody was used to detect protein expression levels in serpin-treated cells compared to control-treated cells. Relative protein expression was quantified by normalizing to b-tubulin expression. The inset shows a representative Western blot. (E) Huh-7.5.1 cells were treated with DMSO (control) or with 10 mM PF-429242 for 24 hours, the compound was removed, and the cell lysates were harvested after an additional 48 hours. Relative ADRP expression (normalized to b-tubulin) in inhibitor-treated cells compared to control cells was quantified by subjecting total cell lysates to Western blot analysis. Values are plotted relative to protein expression in control cells, which are set to 1. Results (mean 6 SEM) from 3 independent experiments are shown. *p,0.05. doi:10.1371/journal.ppat.1002468.g004 inhibiting HCV infection (JFH-1 strain) of Huh-7.5.1 cells in a dosedependent manner. Furthermore, we reveal the common molecular and cellular mechanisms of action of the SKI-1/S1P inhibitors and demonstrate that they act as negative modulators of cytoplasmic LD abundance (Figure 7) , an organelle central to HCV assembly [8] and liver steatosis. To investigate the biological consequences of inhibiting SKI-1/ S1P endoprotease activity on biochemical pathways of lipid homeostasis hijacked by HCV, we first employed a protein-based inhibitor strategy. We engineered and developed a novel, recombinant adenovirus expressing an effective and specific secretory pathway, SKI-1/S1P-directed serpin, Spn4A.RRLL(s). We showed that Spn4A.RRLL(s) forms a kinetically trapped heatand SDS-stable complex with SKI-1/S1P molecules characteristic of other physiological serpin-protease pairs. We then demonstrated that it blocks the SKI-1/S1P-mediated cleavage of endogenous SREBP-1 and the expression of SREBP downstream effector gene products (e.g., SREBP-2, LDLR, and PCSK9). The SREBP target gene products identified in Spn4A.RRLL(s)-treated cells are involved in lipid homeostasis and reported to participate in HCV-host interactions. SREBP-2, along with the other SREBP isoforms SREBP-1a and -1c, are activated by HCV-encoded proteins or during HCV infection [19, 31] . PCSK9 has been implicated in HCV infection through its regulation of two HCV entry factors: CD81 and LDLR [68] . Although the specific role of LDLR in HCV infection is unclear, increasing evidence indicates that LDLR promotes attachment and uptake of lipoproteinassociated HCV particles into hepatocytes [29, 69, 79, 80] . In addition to blocking SREBP-mediated up-regulation of hepatic genes during HCV infection, we hypothesized that blocking SKI-1/S1P endoprotease activity would also compromise cellular lipid storage. Analysis of intracellular lipid content and cytoplasmic LD abundance in Spn4A.RRLL(s)-expressing cells confirmed our hypothesis. In addition, we observed a decrease in ADRP/perilipin 2 abundance in Spn4A.RRLL(s)-expressing cells compared to control-treated cells. The physiological role of ADRP has yet to be fully established but it plays an important role in LD structure and formation. Interestingly, ADRP is degraded through the proteasome-dependent pathway during regression of lipidstoring cells, indicating that when ADRP is not bound to LDs, such as in SKI-1/S1P-inhibited cells, it will be susceptible to rapid proteasomal degradation [81] . Clinical studies have demonstrated that there is a correlation between the level of ADRP and the degree of hepatocyte steatogenesis in humans [82] . In addition, ADRP is found to be up-regulated in fatty liver in humans and in mice with liver steatosis [83] . Collectively, these observations suggest that inhibition of SKI-1/S1P offers an attractive therapeutic target for reducing HCV-induced liver steatosis. As hypothesized, inhibiting SKI-1/S1P-mediated SREBP endoproteolytic cleavage events using Spn4A.RRLL(s) resulted in a dose-dependent decrease in HCV infection. We demonstrated that HCV core expression and HCV RNA levels are reduced in Spn4A.RRLL(s)-expressing cells following HCV infection, leading to a robust reduction in extracellular infectious HCV particle release. Western blotting also confirmed that HCV core protein post-translational processing was unaltered by serpin expression because the size of the protein was unaltered. Supplementing 1) The inactive SKI-1/S1P zymogen is biosynthesized in the ER and traffics to the Golgi apparatus following intramolecular autocatalytic maturation of the proenzyme [35, 46, 99, 100] . (2) During HCV infection, the SREBP pathway is activated by a variety of molecular mechanisms [19, [30] [31] [32] . (3) For SREBP to activate genes involved in lipid biosynthesis, its N-terminal domain must be released through sequential endoproteolytic cleavage first by SKI-1/S1P and then by S2P [37, 38] . (4) The released N-terminal domain translocates to the nucleus and activates various aspects of lipid metabolism [36] . (5) Activation of lipid biosynthesis increases LD formation where the HCV core protein localizes to orchestrate HCV assembly and subsequent secretion [8, 10, 16] . (6) Biosynthesis of LDLR, a proposed receptor for HCV entry, is also activated by SREBP signaling [69, 79, 101] . (7) Spn4A.RRLL(s) is a secretory pathwayexpressed serpin (Figure 1 ). (8) Spn4A.RRLL(s) interacts and forms a covalent complex with enzymatically active SKI-1/S1P molecules (Figure 2 and S3) in the Golgi apparatus preventing SKI-1/S1P-mediated endoproteolytic cleavage of SREBP protein ( Figure 3A ). (9) A small-molecule inhibitor PF-429242 also efficiently inhibits SKI-1/S1P endoproteolytic activity ( Figure 3A ). (10) SKI-1/S1P inhibition blocks expression of the putative HCV receptor, LDLR ( Figure 3B and S4) , and reduces HCV entry ( Figure 5) . (11) The expression of other SREBP-regulated genes, such as PCSK9 and SREBP-2, are also blocked ( Figure 3B ). (12) Downstream lipid synthesis is interrupted resulting in overall reduced intracellular cholesterol-ester and triglyceride abundance ( Figure 3C and 3D) . (12) This is then detected as a decrease in LD abundance (Figure 4) , which impedes assembly and secretion of infectious HCV particles. doi:10.1371/journal.ppat.1002468.g007 Spn4A.RRLL(s)-expressing cells with compounds such as mevalonate, oleate, and cholesterol resulted in an incomplete rescue of HCV infection, suggesting that the antiviral activity of our proteinbased inhibitor cannot be explained solely by the decreased availability of lipids in these cells. This also supports our hypothesis that reduced LDLR levels may compromise HCV entry into Spn4A.RRLL(s)-treated hepatocytes. LDLR, a well-established SREBP-regulated gene, has been repeatedly shown to support HCV entry into hepatocytes [69, 80] . Also, no significant reductions in HCV RNA levels were observed in Spn4A. RRLL(s)-treated cells harbouring subgenomic HCV replicons or following full-length genomic HCV RNA transfection in Huh-7.5.1 cells. Altogether, these studies indicate that the observed decline in HCV RNA levels following HCV infection may result from compromised HCV entry. To gain further insight into the different stages of the viral lifecycle targeted by our SKI-1/S1P inhibitor, we used an activesite-directed small-molecule inhibitor of SKI-1/S1P, PF-429242. This pharmacologic inhibitor of SKI-1/S1P has recently been characterized for its effectiveness in blocking cleavage of SREBP-2, for blocking expression of SREBP-activated genes, and also for inhibiting arenavirus glycoprotein processing [64, 84] . In contrast to our recombinant adenovirus-expressed serpins, PF-429242 can be added extracellularly to rapidly inhibit SKI-1/S1P. This allows us to study the biological impact of blocking SKI-1/S1Pdependent pathways during both early and late stages of HCV infection. We first confirmed a reduction in abundance of neutral lipids and ADRP/perilipin 2 expression in PF-429242-treated cells. Then, we confirmed that inhibition of SKI-1/S1P using PF-429242 blocks HCV infection and extracellular infectious virus production in a dose-dependent manner. The anti-HCV activity of PF-429242 is very robust and particularly striking. A single 24hour pre-treatment with the compound was sufficient to block HCV infection, and the antiviral effect of PF-429242 was still apparent 72 hours post-treatment. Importantly, pharmacological treatment of already infected HCV cells resulted in a 90% reduction of HCV virus production. Similar to Spn4A.RRLL(s) treatment, PF-429242 did not reduce HCV RNA levels in the two stable subgenomic replicon cell lines that were examined. This, in combination with the observed reduction in abundance of a central organelle (LD) involved in HCV assembly, and the reduction in HCV particle secretion in cells treated 24 hours after HCV inoculation, supports PF-429242 as an inhibitor of late stages of the HCV lifecycle, i.e., during assembly or egress. Taken all together, these results indicate that inhibiting SKI-1/S1P can interrupt the HCV lifecycle at multiple stages of viral infection both preventing naïve cells from becoming infected and preventing virus release from already infected cell populations. Thus, developing more effective active-site-directed SKI-1/S1P small-molecule inhibitors (, nM range) with better pharmacokinetic properties [64] could lead to novel indirect-acting antiviral treatment options for HCV-infected patients [42, 76, 85, 86] . Importantly, inhibiting the SREBP pathway in HCV-infected cells, which have exacerbated lipid production and which are steatotic, may relieve symptoms caused by chronic HCV infection in addition to blocking viral infection [87] . In conclusion, pharmacologic inhibition of SKI-1/S1P offers a very promising avenue for the development of novel anti-HCV therapeutics (Figure 7) . On one hand, targeting a host cell master molecular switch such as SKI-1/S1P with a novel class of drugs compromising multiple stages of the virus lifecycle would have the main advantage of making it more difficult for the virus to develop escape mutations [86, 88] . On the other hand, the toxicity issues associated with the inhibition of host cell proteases such as SKI-1/ S1P [89] could be addressed by using adjunctive therapy, combining our novel class of lipid-modulating agents with the current standard of care or with the appropriate synergistic directacting antivirals [85, 86] . Finally, our results reveal that targeting host LD biogenesis by inhibiting SKI-1/S1P endoproteolytic activity may have farreaching applications in the therapeutic treatment of other important human Flaviviridae viruses such as dengue virus, whose replication and pathogenesis also depend on the interaction with lipid droplets [7] . A plasmid containing the cDNA of an HCV consensus clone isolated from a Japanese patient with fulminant hepatitis (JFH-1) (GenBank accession number AB047639) [91] cloned behind a T7 promoter (pJFH-1; a generous gift from Dr. Takaji Wakita, National Institute of Infectious Diseases, Tokyo, Japan) was used to generate genomic HCV RNA and infectious HCV stocks as previously described in [61] . Purified HCV RNA was used to transfect Huh-7.5.1 cells as a means of studying HCV infection independently of receptormediated entry. Five micrograms of purified RNA was incubated with 10 ml of lipofectamine 2000 (Invitrogen) in minimal essential media (MEM) for 30 minutes. The RNA-lipid complexes were added to cells in MEM for 16 hours; then cells were washed with phosphate-buffered saline (PBS) and complete media was added for the remainder of the experiment. The amount of infectious HCV particles generated for viral stocks or in the described experiments was determined using a modified, previously described protocol [61] . Briefly, 1610 4 Huh-7.5.1 cells were plated in each well of a 96-well plate and infected with 10-fold serial dilutions of HCV-infected cell media. At 72 hours post-infection, cells were fixed and probed as described in the ArrayScan Quantification methods section. An ArrayScan VTI High Content Screening (HCS) Reader (Thermo Scientific) was used to acquire images of the entire group of infected wells. Titers were determined by manually counting foci (fluorescence forming units (FFU)) in the lowest dilutions with positive signal. In silico homology model of Spn4A.RRLL variant The Drosophila melanogaster Spn4B sequence (GeneBank Accession number gi|24586105|ref|NP_524955.2) exhibits 34% sequence homology with the human neuroserpin (hNS), for which a crystal structure is available [94] in the Protein Data Bank (PDB ID: 3F5N). Of the five chains in this pentameric structure of hNS, chain B is most well resolved with the fewest missing residues, and it was used as the template for the homology model presented in Figure 1B . The model was built and refined using the SwissPDB Viewer. The C-alpha residues in this model structure align to 1.9 Å RMSD with reference to the hNS structure. The first 997 amino acids of human SKI-1/S1P lacking the Cterminal transmembrane domain but containing a C-terminal 8-His-tag (PGDDDDKHHHHHHHHSGS) were expressed in Sf9 insect cells as previously described [47] . Two liters of cell culture supernatant were used for purification. Two hundred milliliters of 200 mM Tris/HCl pH 8.0, 500 mM NaCl was added, and then the pH was adjusted to pH 8.0 by further addition of 2 M NaOH. The resulting precipitate was removed by centrifugation at 10000 x g for 30 minutes and subsequently filtered through a glass filter. The cleared supernatant was then applied to a small (0.9 ml column volume) IMAC column (Ni-Sepharose, GE Healthcare, Freiburg, Germany) by continuous flow (1.0 ml/minutes). The column had previously been equilibrated in 50 mM Tris/HCl pH 8.0, 500 mM NaCl (buffer A), and bound recombinant SKI-1/S1P was eluted with a continuous gradient over 30 column volumes to buffer A plus 300 mM imidazole. Collected fractions were assayed for SKI-1/S1P enzymatic activity as previously described [47] using the paranitroanilide (p-NA) acetylated (Ac) tetrapeptidyl substrate Ac-RRLL-pNA [custom synthesized by Peptides International (Louisville, Kentucky, USA)]. The most active fractions were pooled. Concentration and buffer exchange to buffer A was then done using spin concentrators (Millipore, Billerica, MA, USA) with a molecular weight cutoff of 30 kDa. The final preparation was, after addition of 30% v/v glycerol, stored at -80uC and had a specific activity of 0.018 U/mg (measured as above). Recombinant His-tagged furin (0.432 mg/ ml) was purchased from R & D Systems (Minneapolis, MN, USA), and reactions with adenovirus recombinant serpin were performed under the buffer conditions for furin assays as previously described [54, 58] . Cultured cells were washed with ice-cold PBS and re-suspended in cold radioimmunoprecipitation assay (RIPA) buffer (50 mM Tris-HCl pH 8, 150 mM NaCl, 1% octylphenyl-polyethylene glycol [IGEPAL], 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulfate [SDS] containing 1 X Complete, EDTA-free, protease inhibitor cocktail [Roche, Laval, QC, Canada]). Whole cell extracts were vortexed and then clarified by centrifugation at 12000 x g for 15 minutes. Soluble extracts mixed with 2 X sample buffer (62.5 mM Tris-HCl, pH 6.8, 25% glycerol, 2% SDS, 0.01% bromophenol blue, and 5% beta-mercaptoethanol) were electrophoresed on 8-15% SDS polyacrylamide gels and transferred to nitrocellulose membranes. Membranes were blocked in Odyssey blocking buffer (LI-COR Biosciences) for one hour, and proteins of interest were detected by probing with the appropriate primary and secondary antibodies diluted in Odyssey blocking buffer containing 0.1% Tween 20. Protein bands were detected and quantified using the Odyssey Infrared Imaging System (LI-COR Biosciences). All immunoblots were scanned at a wavelength of 700 nm for detecting IRDye 680 labeled antibodies and at a wavelength of 800 nm for IRDye 800CW conjugated antibodies [95, 96] . Signal intensities were quantified by means of the Odyssey software version 3.0. Beta-tubulin was always used as a loading control and for normalizing protein expression. Media samples analyzed for secreted Spn4A variants were taken directly from cultured cells, mixed with 2 X sample loading buffer, and subjected to the described Western blot analysis. [47] ; furin buffer contains 100 mM HEPES, pH 7.5, 1 mM CaCl 2 , 0.5% Triton X-100 [54, 58] , 1 X Complete EDTA-free protease inhibitor cocktail), and 11.6 ng/ml SKI-1/ S1P or 2.4 ng/ml furin. The enzyme mixture was incubated at 30uC for 30 minutes, and the reaction was stopped with 12.5 mM EDTA (final concentration). After completion, products were resolved on a 10% SDS-gel. The high-molecular weight band (EI) was visualized as described above for Western blotting. In black flat-bottom 96-well plates (BD Biosciences), cells were plated (.10,000 cells/well) and infected as described in the methods below. Following infection, cells were fixed in 4% formaldehyde v/v diluted in PBS and blocked in PBS containing 3% BSA, 0.3% Triton X-100, and 10% FBS. Cells were first probed with HCV anti-core antibody (1:500) in PBS containing 3% BSA and 0.3% Triton X-100 (Binding Buffer), then incubated with Alexa Fluor-568-conjugated donkey anti-mouse secondary antibody (1:1000) and 10 mg/ml Hoechst dye. Cells were analyzed by a quantitative, high-throughput, fluorescence microscope system called the Cellomics ArrayScan VTI High Content Screening (HCS) Reader (Thermo Scientific) using the software Target Activation BioApplication (TABA). TABA was used to count the total number of cells (Hoechst-stained nuclei) and the percentage of those cells that were expressing HCV core (positive signal at 568 nm wavelength). . Nuclear fractionation of cell lysates was performed at 4uC as described previously with modifications [64, 66] . Cells were harvested in 400 ml buffer C (10 mM HEPES/KOH, pH 7.6, 10 mM KCl, 1.5 mM MgCl 2 , 1 mM EDTA, 1 mM EGTA, 250 mM sucrose) containing 1 X complete protease inhibitor cocktail (Roche). To shear DNA, the cells were passed 20 times through a 23-gauge needle. The lysate was centrifuged at 1100 x g for 7 minutes and the resulting supernatant was centrifuged again at 25000 x g for 60 minutes to obtain the membrane pellet. The pellet containing membranebound SREBP-1 was re-suspended in 75 ml of SDS-lysis buffer (10 mM Tris HCl, 100 mM NaCl, 1% SDS, 1 mM EDTA, 1 mM EGTA, pH 6.8). The pellet from the 1100 x g spin was resuspended in 100 ml buffer D (20 mM HEPES/KOH, 420 mM NaCl, 1.5 mM MgCl 2 , 2.5% glycerol, 1 mM EDTA, 1 mM EGTA, pH 7.6) containing 1 X complete protease inhibitor cocktail. Nuclear pellets were rocked for 1 hour, after which the samples were centrifuged at 25000 x g for 60 minutes to obtain the clarified supernatant containing the nuclear fraction. After Huh-7.5.1 cells were seeded onto coverslips for 24 hours, they were infected with adenovirus (moi 50) for 72 hours. Cells were fixed in 4% v/v formaldehyde in PBS, then permeabilized and blocked in PBS containing 0.05% saponin (wash buffer) and 1% BSA (binding buffer). Blocking of cells stained with BODIPY 493/503 was done in the presence of 0.2 M glycine to reduce background fluorescence. Cells were probed with primary antibodies in binding buffer, then incubated with a secondary antibody, Hoechst dye (10.0 mg/ml), and BODIPY 493/503 (1.0 mg/ml; when indicated) diluted in PBS. Cells were mounted onto slides with an anti-fade solution and sealed with clear nail polish. The slides were then imaged using a Leica TCS SP5 confocal microscope (Leica Microsystems, Wetzlar, Germany) or an Olympus Fluoview FV1000 laser scanning confocal microscope (Olympus Corporation, Tokyo, Japan) [95] [96] [97] . Leica MM AF Software (Leica Microsystems) was used to count the number of LDs (green channel) in cells expressing Spn4A.RRLL(r) (n = 21) or Spn4A.RRLL(s) (n = 15) (cells positive in the red channel). LDs in cells treated with Ad-Empty (n = 23) were also enumerated. All quantified images were acquired using the same laser intensity and gain settings, and LDs were enumerated by applying the same threshold setting to each image. Huh-7.5.1 cells were infected with recombinant adenovirus at different moi in complete media with or without exogenous sterols (50 mM sodium mevalonate, 20 mM sodium oleate, 5.0 mg/ml cholesterol). After 48 hours, the cells were infected with HCV moi 0.1 or transfected with purified HCV genomic RNA for 72 hours, and then cells were analyzed by Cellomics ArrayScan HCS, or total RNA was isolated from cell extracts using the RNeasy plus kit (Qiagen, Mississauga, ON, Canada) including on-column DNase digestion. Media from treated and infected cells were harvested for HCV titer determination as described above. To examine the Spn4A.RRLL(s) mediated block in PCSK9, LDLR, and SREBP-2 expression, Huh-7.5.1 cells were grown in media supplemented with LPDS for 24 hours, infected with adenovirus variants, and harvested 72 hours later. Purified total RNA was reverse transcribed to cDNA using TaqMan reverse transcription reagents (random hexamers; Applied Biosystems, Foster City, CA, USA). Real-time quantitative PCR was carried out using Brilliant II Fast QPCR reagents (Stratagene, La Jolla, CA, USA) according to the manufacturer's instructions on an Mx3005P QPCR system (Stratagene). Online ProbeFinder software (Roche Applied Science) was used to find primers that would allow amplification of the HCV RNA 59 end in combination with the Human Universal Probe Library from Roche (Roche Applied Science). For amplification of the HCV RNA 59 region, 400 nM of both forward primer (59-CAT-GGCGTTAGTATGAGTGTCG-39) and reverse primer (59-GG-TTCCGCAGACCACTAT-39) were used in combination with 200 nM of probe #75 from the Human Probe Library (Roche). HCV RNA levels were relatively quantified across samples and normalized to beta-actin RNA levels using 500 nM primers (forward: 59-GCC CTG AGG CAC TCT TCC and reverse: 59 GGA TGT CCA CGT CAC ACT TC-39) and 250 nM probe (59AC TCC ATG CCC AGG AAG GAA GGC-39 with a 59 Cy5 fluorophore and 39 black hole quencher). Cell viability was determined using CellTiter 96 AQueous One Solution Cell Proliferation Assay (Promega, Madison, WI, USA). This assay employs a tetrazolium compound [3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium, inner salt; MTS], which is bio-reduced by cells into a colored formazan product that can be detected in tissue culture media at 490 nm wavelength [98] . PF-429242, an active site-directed small-molecule inhibitor of SKI-1/S1P [64, 84] , was synthesized by Dr. Peter Chua at the Center for Drug Research and Development (CDRD) at the University of British Columbia (Vancouver, BC, Canada) according to previously described protocols [65] . To investigate the antiviral activity of the small molecule, Huh-7.5.1 cells were treated with PF-429242 for 24 hours. After 24 hours of treatment, the media was removed and then cells were infected with HCV (moi 0.1) for 48 or 72 hours. Alternatively, cells were first infected with HCV for 24 hours; then, the media was removed and replaced with media containing various concentrations of PF-429242 for a further 48 hours. Intracellular HCV infection levels were determined using Cellomics HCS ArrayScan. Infectious extracellular titers were determined in media of 72-hour HCVinfected cells. To measure the level of intracellular lipids following 72-hour recombinant adenovirus expression or 48-hour PF-429242 treatment, cellular extracts were harvested in 1% triton-X 100 in PBS (for phospholipid, cholesterol, and protein assay) or 5% triton-X 100 in H 2 O (for triglyceride assay). To extract triglycerides, samples were slowly heated to 90uC and brought to room temperature, twice. Total cholesterol, cholesterol esters (Amplex Red Cholesterol Assay kit, Invitrogen), phospholipids, and triglycerides (EnzChrom, BioAssay Systems, Hayward, CA, USA) were quantified using commercially available kits. Lipid levels were normalized to cellular protein content (DC Protein Assay, Bio-Rad, Hercules, CA, USA). The sigmoidal fit function in Igor Pro software (WaveMetrics, Inc., Portland, OR, USA) was used for fitting HCV and PF-429242 inhibition curves and for determining EC 50 values. The reported EC 50 values are the average of the values calculated from three independent experiments plus or minus the standard deviation. The student's t-test (unpaired) was used to calculate significance, which is represented in the figures by the following notation: * denotes p,0.05, ** denotes p,0.01, and *** denotes p,0.005. Figure S1 Optimization of adenovirus-expressed Spn4A.RRLL(r) expression in Huh-7.5.1 cells. Huh-7.5.1 cells were infected with moi 1, 12.5, 25, and 50 of the intracellularly retained serpin Ad-Spn4A.RRLL(r). Treated cells were fixed 48 hours post-infection and probed for serpin expression using mouse anti-FLAG antibody. Cell nuclei were stained with Hoechst dye to determine the total cell number. The percentage of Spn4A.RRLL(r)-expressing cells was quantified using Cellomics HCS. Results (mean 6 SEM) from 2 independent experiments are shown. (TIF) Figure S2 Effect of Spn4A variant treatment and HCV infection on Huh-7.5.1 cell growth. Huh-7.5.1 cells were infected with various moi (1 -50) of Ad-Empty, Ad-Spn4A.RRLL(r), or Ad-Spn4A.RRLL(s) for 48 hours in complete media. Treated cells were infected with HCV (moi 0.1) and fixed 72 hours post-infection. Fixed cells were probed with Hoechst dye to stain for cell nuclei, which were then quantified using Cellomics HCS to determine the relative number of cells in each well under the varying conditions. All values are expressed as relative cell number in serpin-treated cells compared to cells infected with Ad-Empty, which is set to 1. Results (mean 6 SEM) from 3 independent experiments are shown. (TIF) Figure S3 Serpin-like properties of recombinant adenovirus-expressed Spn4A variants expressed in Huh-7.5.1 cells. Huh-7.5.1 cells were infected with recombinant adenovirus expressing the His-and FLAG-tagged Spn4A variants indicated or the Ad-Empty control for 72 hours. Media alone (upper panels) or cell extracts (lower panels) lysed in RIPA buffer were combined with recombinant His-tagged SKI-1/S1P [47] or His-tagged furin for 30 minutes at 30uC. Samples were prepared for Western blot analysis and probed with rabbit anti-FLAG antibody to detect SDS-and heat-stable protease-serpin complex formation and also to distinguish serpin bands from protease bands on the Western blots. All Western blots shown are representative of at least 2 independent experiments. (TIF) Figure S4 Time course analysis of LDLR expression in Spn4A.RRLL(s)-treated cells. Huh-7.5.1 cells were grown in complete media for 24 hours and then infected with Ad-Empty (control) or Ad-Spn4A.RRLL(s). Cell extracts were harvested for Western blot 24, 48, and 72 hours post-infection, and lysates were then subjected to Western blot. Anti-LDLR antibody was used to detect protein-expression levels in control-and Ad-Spn4A. RRLL(s)-treated cells, and b-tubulin was probed for normalizing LDLR expression. Values are plotted relative to LDLR expression in control (Ad-Empty)-treated cells, which was set to 1. The representative results of 2 independent experiments are shown. (TIF) Figure S5 Spn4A.RRLL(s) does not block HCV core production post-transfection of HCV RNA. Huh-7.5.1 cells were infected with Ad-Empty (control), Ad-Spn4A.RRLL(r), or Ad-Spn4A.RRLL(s) (moi 50) for 48 hours in complete media and then transfected with genomic HCV RNA for 72 hours. Relative HCV-core expression (normalized to b-tubulin) in serpin-treated cells compared to control-treated cells was quantified by examining total cell lysates using Western blot analysis. Results (mean 6 SEM) from 2 independent experiments are shown. A representative Western blot is shown to the right of the graph. (TIF) Figure S6 The effect of PF-429242 on cell viability. Huh-7.5.1 cells were treated with DMSO (control) or various concentrations of PF-429242 for 24 hours before the inhibitor was removed, and fresh complete media was added to the cells for an additional 48 hours. The relative cytotoxicity of the compound was then determined using an MTS-based cell viability assay. The absorbance measured at 490 nm is proportional to the number of living cultured cells. Results (mean 6 SEM) from 3 independent experiments are shown. Statistical significance was calculated for PF-429242-treated cells compared to DMSO-treated cells. (TIF)

HMGB1, an alarmin promoting HIV dissemination and latency in dendritic cells

Dendritic cells (DCs) initiate immune responses by transporting antigens and migrating to lymphoid tissues to initiate T-cell responses. DCs are located in the mucosal surfaces that are involved in human immunodeficiency virus (HIV) transmission and they are probably among the earliest targets of HIV-1 infection. DCs have an important role in viral transmission and dissemination, and HIV-1 has evolved different strategies to evade DC antiviral activity. High mobility group box 1 (HMGB1) is a DNA-binding nuclear protein that can act as an alarmin, a danger signal to alert the innate immune system for the initiation of host defense. It is the prototypic damage-associated molecular pattern molecule, and it can be secreted by innate cells, including DCs and natural killer (NK) cells. The fate of DCs is dependent on a cognate interaction with NK cells, which involves HMGB1 expressed at NK–DC synapse. HMGB1 is essential for DC maturation, migration to lymphoid tissues and functional type-1 polarization of naïve T cells. This review highlights the latest advances in our understanding of the impact of HIV on the interactions between HMGB1 and DCs, focusing on the mechanisms of HMGB1-dependent viral dissemination and persistence in DCs, and discussing the consequences on antiviral innate immunity, immune activation and HIV pathogenesis.

High mobility group box 1 (HMGB1) is a nuclear DNAbinding protein actively released by innate immune cells in response to exogenous pathogen-derived molecules, acting as a danger signal and triggering inflammation. HMGB1 is a proinflammatory cytokine that is essential for maturation of dendritic cells (DCs), their migration to lymphoid tissues and Th1 polarization of naïve T cells. HMGB1 signals by binding to Toll-like receptor 4 (TLR4) to activate MyD88-dependent nuclear translocation of NF-kB, which upregulates the expression and release of cytokines and other inflammatory mediators. HMGB1 is expressed at the synapse between NK cells and DCs, it is pivotal during NK-DC cross talk, promoting DC maturation and protecting them from lysis. HMGB1 triggers human immunodeficiency virus (HIV) replication in latently infected primary myeloid cells. What are the molecular mechanisms involved in vivo in the disruption of NK cell-DC cross talk during chronic HIV-1 infection, and what are the consequences on both NK cell killing activity and DC-dependent promotion of adaptive immune responses? Does HMGB1 has a role in the trans-infection of T lymphocytes with HIV-1 through the exosome-dissemination pathway? Given that HMGB1 can combine with LPS to trigger TLRs, and TLR-mediated immune activation results in the production of proinflammatory cytokines, to what extent does HMGB1 contribute to generalized immune activation and disease progression in HIV-1-infected individuals? What is the contribution of HMGB1 to HIV dissemination and the establishment of HIV reservoirs in DCs? Would the specific targeting of c-FLIP or c-IAPs in DCs contribute to the depletion of HIV-1 reservoirs? Given the expression of HMGB1 and its receptor RAGE in active neurological diseases, including multiple sclerosis and Alzheimer's disease, does it has a role in HIV-associated neurological disorder? High-mobility group box 1 protein (HMGB1) (also known as amphoterin or HMG1) was originally defined as a non-histone nucleosomal protein that is important for the regulation of transcription. It is a 215 amino-acid protein, encoded on chromosome 13q12, which is highly conserved between species (99% species homology between rodents and humans). HMGB1 contains two internal repeats of positively charged domains, the A-and B-Box, in the N terminus, and a negatively charged COOH terminus ( Figure 1 ). The two boxes bind to the minor groove of chromatin, thus modifying DNA architecture. 1 This facilitates the binding of regulatory protein complexes to DNA such as V(D)J recombinases 2 and p53-p73 transcriptional complexes. [3] [4] [5] In its resting state, the acidic tail of HMGB1 interacts with specific residues in the A-Box and B-Box, forming an extended and flexible segment, shielding them from other interactions that might occur before HMGB1 binds DNA. 6 HMGB1 likely has a role in DNA repair and replication. HMGB1 overexpression, which is observed in many tumors, accelerates cell cycle progression, and recent data suggest that endogenous HMGB1 is a critical pro-autophagic protein that enhances cell survival 7 and that HMGB1-induced autophagy promotes chemotherapy resistance in leukemia cells. 8 The discovery by Kevin J Tracey et al. (1999) in a mouse model of endotoxaemia that lipopolysaccharide (LPS)activated macrophages release HMGB1, but later than secretion of the pro-inflammatory cytokines TNF-a and interleukin 1 (IL-1), and that protection against endotoxin lethality could be obtained by administration of anti-HMGB1 antibodies 9 has revealed that HMGB1 is a proinflammatory mediator able to alert the immune system to tissue damage and to trigger an immediate response. The term 'alarmin' has been proposed to differentiate the endogenous molecules that are very rapidly released or produced in response to microbial infection or tissue injury, and act as potent effectors of innate defense. 10 Alarmins have antimicrobial, enzymatic or chromatin-binding activities and they share common features, including their rapid passive release from necrotic cells or secretion from cells of the innate immune system (macrophages, natural killer (NK) cells) in response to infection, they bind to TLRs and receptors of antigen-presenting cells such as DCs, thus promoting adaptive immunity, and they are involved in the reconstruction of tissues destroyed secondary to inflammation. 11 Based on these criteria, a list of putative alarmins has been proposed, including the defensins, eosinophil-derived neurotoxin, thymosins, annexins, HSPs, or IL-1a. 12 HMGB1 remarkably fulfills these criteria and it is probably the best-characterized alarmin. The crucial role of HMGB1 not only in response to infection, injury and inflammation, but also its pathological effects in many diseases have recently challenged important questions regarding its biological activities and pathological effects. Recent studies have established the involvement of HMGB1 in not only acute and chronic inflammatory conditions, including sepsis, 13 rheumatic diseases, 14 or SLE, 15 but also viral infectious diseases, such as that induced by SARS, hepatitis viruses, influenza viruses 16 or HIV. 17 It is currently unclear whether HMGB1-mediated inflammatory response contributes to the pathogenesis of various viral diseases, but it has been suggested to be involved in SARS-associated injurious pulmonary inflammatory response, persistent liver injury in hepatitis patients, or pathogenesis of West Nile encephalitis. 18 Regarding HIV infection, elevated plasma levels of HMGB1 were detected during progressive HIV-1 infection, positively associated with viral replication. 17 In vitro, HMGB1 may trigger or inhibit HIV-1 replication, depending on the target cell and the microenvironment. 19, 20 Recent studies analyzed the impact of HMGB1 on the fate of HIV-1-infected DCs, and the data suggested not only a possible contribution of this protein to the functional impairment of DCs but also to HIV dissemination and persistence. 21, 22 This review will discuss the mechanisms whereby HMGB1 contributes to innate immunity by regulating maturation and functions of DCs but also how it may contribute to viral latency and HIV disease pathogenesis. HMGB1, a DAMP that Likes DCs HMGB1, a sentinel for nucleic-acid-mediated response in DCs. During microbial infection, the activation of innate immune responses by DNA and RNA is essential to protective immune responses and is mediated by the NH2 COOH A box B box Acidic tail 1 79 89 163 186 215 DNA binding domain DNA binding domain 150 183 RAGE-binding domain 28 44 NLS1 179 185 NLS2 C106 TLR4 binding site Proinflammatory cytokine-domain 89 108 100 189 202 transmembrane TLRs and cytosolic receptors. 23 TLRs belong to a family of pattern recognition receptors (PRRs) that have essential roles in innate immunity. They are a class of single-membrane-spanning receptors that have the ability to recognize structurally conserved molecules from bacteria. Engagement of TLRs activates the immune response. HMGB proteins bind to all immunogenic nucleic acids and have recently been identified to serve as universal sentinels for nucleic-acid-mediated innate immune responses. 24 Activation of macrophages or DCs with microbial cytosinephosphate-guanosine (CpG)-DNA, resulting in secretion of proinflammatory cytokines, involves TLR9 initially localized in the endoplasmic reticulum (ER). The mechanism of TLR activation by CpG-DNA was recently discovered and HMGB1 was found to have a crucial role. HMGB1 binds to CpG-DNA and receptor for advanced glycated endproducts (RAGE), the first identified receptor for HMGB1, and this DNA-protein complex preassociates with TLR9 in the ER-Golgi intermediate compartment, thus accelerating the delivery of microbial DNA to TLR9 and leading to the recruitment of the TLR adaptor molecule MyD88. 25 Thus, HMGB1 and RAGE are pivotal for TLR9-dependent induction of genes encoding type I interferon after stimulation of DCs with DNA-containing complexes. Ablation or depletion of HMGB1 impairs redistribution of TLR9 to early endosomes in response to CpG-oligodeoxynucleotides (ODN), leading to a decreased response to CpG-ODN, but which could be complemented by extracellular HMGB1. 25 In addition, HMGB1 was found to be a key factor in immune complex-triggered activation of autoreactive B cells and the induction of type I interferon by plasmacytoid DCs, thus possibly contributing to SLE. 25 HMGB1, a cytokine that initiates host defense. HMGB1 is a nuclear protein and, to function as a cytokine, it must be released in the extracellular milieu. This occurs either via passive release from necrotic cells, 26, 27 or by active secretion by cells of the innate immune system. Wang et al. 9 first reported that HMGB1 was liberated from macrophages stimulated with LPS, and that HMGB1 had an important role in experimental sepsis. HMGB1 secretion from LPS-primed macrophages was shown to require the inflammasome components apoptotic speck protein containing a caspase recruitment domain (ASC), caspase 1 and NALP3. 28 Monocytes, macrophages and immature DCs (mDCs) secrete HMGB1 in response to LPS, TNF-a, or IL-1b stimulation. 29 IFN-g can induce HMGB1 release from macrophages that, at least in part, requires induction and signaling through TNF-a. 30 Secretion of a nuclear protein requires a tightly controlled relocation program. Several forms of post-translational modifications, such as acetylation, phosphorylation and oxidation result in the accumulation of HMGB1 in the cytosol. [31] [32] [33] [34] Upon activation with LPS, monocytes and macrophages acetylate HMGB1 extensively, allowing its relocalization from the nucleus to the cytosol, and further concentration into secretory lysosomes. 35 Thus, because it does not contain a leader sequence, HMGB1 is secreted via a non-classical vesiclemediated secretory pathway. A second pathway of HMGB1 nuclear/cytoplasmic shuttling has been reported involving phosphorylation, 33 mediated by protein kinase C. 36 In neutrophils, HMGB1 is post-translationally methylated, which alters its conformation and weakens its DNA-binding activity, causing its cytoplasmic localization. 34 Recently, HMGB1 has been shown to undergo oxidation that may have the potential to modulate various aspects of its function, including subcellular localization, interaction with DNA, cytokine activity and proinflammatory activity. 37 Indeed, induction of immunological tolerance by apoptotic cells was shown to require caspase-dependent ROS production by mitochondria, which oxidized the danger signal HMGB1, and thereby neutralized its damage-associated molecular pattern molecule (DAMP) function, including the ability to activate DCs. 38 From the reverse perspective, in necrotic cells that do not generate ROS and provide fully active HMGB1, treatment with H 2 O 2 inactivates the DAMP function of the protein. Thus, DAMP is released and can work, but briefly and in a short range, until it gets oxidized. This restricts its activity temporally and spatially, preventing a prolonged stimulation of its targets and thereby limiting inflammation. 39 HMGB1 is essential for DC maturation, migration and Th1 polarization. Inflammatory signals activate antigenpresenting DCs, which undergo a differentiation process referred to as maturation and migrate to secondary lymphoid organs. HMGB1 has a crucial role in this process, acting as a chemoattractant for immature DCs (iDCs) that involves RAGE, and further inducing DC maturation, as shown by the upregulation of the surface markers CD80, CD83, CD86 and human leukocyte antigen (HLA)-A, B, C, and DC production of cytokines, including IL-6, CXCL8, IL-12p70 and TNF-a. 11, 40 The mobilization of DCs from peripheral tissues is critical for the establishment of T-cell-dependent immune responses or tolerance, because the physical interaction of DCs with naive T cells takes place in the T-cell areas of lymph nodes. Importantly, in vivo homing of DCs to draining lymph nodes depends on RAGE, 41 a finding consistent with the in vitro upregulation of the CCR7 and CXCR4 receptors upon autocrine secretion of HMGB1 by mature myeloid DCs. 40 Thus, the autocrine/paracrine release of HMGB1 and the integrity of HMGB1/RAGE pathway are required for the migratory function of HMGB1. Moreover, HMGB1 secreted by DCs is required for clonal expansion, survival and functional Th1 polarization of naïve T cells, which occurs through the secretion of proinflammatory cytokines, including IL-12, IL-18 and IFN-g 42, 43, 21 In the presence of inhibitors of HMGB1 or of RAGE, DC activated with pathogen-associated molecular patterns (PAMPs) fail to mature 43 ( Figure 2 ). The inflammatory properties of HMGB1 depend on the ability to complex with soluble moieties, including nucleic acids, microbial products (LPS), cytokines (IL-1b) and chemokines. Campana et al. 44 reported that HMGB1 secretion is required for CXCL12 (SDF-1)-dependent migration of DCs. In addition, HMGB1 protects the conformation of CXCL12 in a reducing environment, a state existing in the draining lymph node. However, only a partial inhibition of DC migration was observed in the presence of a CXCL12 inhibitor or HMGB1 A box, suggesting that multiple receptors are responsible for mediating the DC response to the HMGB1/ CXCL12 complex. An interesting study performed in CD24deficient mice revealed how the host distinguishes between danger versus pathogen PAMPs. 45 The proinflammatory activity of HMGB1 is regulated by CD24, which associates with HMGB1, thus inhibiting NF-kappa B activation occurring through CD24 association with Siglec-10. Thus, this CD24-Siglec-10 pathway protects the host against a lethal response to pathological cell death, but it would allow an appropriate response to invading pathogens. 45 DCs, pivotal to adaptive immunity, are early targets of HIV. All lentiviruses can infect macrophage lineage cells in which they generate a persistent infection. HIV-1 has developed a broader tropism leading to preferential infection of CD4 þ T cells, which are progressively destroyed both as a direct viral cytopathic effect and a bystander induction of apoptosis in uninfected cells. 46 Studies in animal models have revealed that critical events to establishing systemic infection take place very quickly at the mucosal portal of entry. Following mucosal exposure to high doses of simian immunodeficiency virus (SIV), the virus can cross the mucosal barrier and it establishes within 3-4 h a small founder population of productively infected cells. 47, 48 Infection then expands locally before virus detection in the draining lymph node, and systemically throughout the secondary lymphoid organs. 49 In vivo, resting CD4 T cells are the initially infected cells in lymphoid tissues. 48 However, the first cells to be infected at the mucosa are the intraepithelial DCs, as shown in vivo in the genital tract of rhesus macaques 1 h after intravaginal inoculation of SIV, and infected DCs reach the draining lymph nodes 18-24 h after SIV exposure, much earlier than CD4 þ T cells. 50 As sentinels, DCs are crucial for the generation of antiviral immunity. They are the most potent antigen-presenting cell in the immune system owing to their superior capacity for acquiring and processing antigens for presentation to T cells and their potential to express high levels of the co-stimulatory molecules that drive T-cell activation and polarization. 51 Thus, they effectively link the innate recognition of viruses to the generation of the appropriate type of adaptive immune response. DCs are a heterogeneous family, including langherans cells in the epidermis, interstitial DCs found in all peripheral tissues, myeloid DCs and plasmacytoid DCs found in the blood. Their heterogeneity resides at several levels, including anatomical location, phenotype and function. 52 DCs express a large repertoire of PRRs and, in response to signals from these receptors, they are activated and migrate to the T-cell area of regional lymph nodes where mDCs present virus-derived epitopes to CD4 þ or CD8 þ T cells. 53 DCs have a prominent role in promoting viral dissemination. Viruses, including HIV, have evolved Figure 2 HMGB1-mediated cross talk between DCs and NK cells is pivotal to DC maturation and further induction of adaptive immunity. The disruption of an epithelial barrier allows invasion of microbial pathogens, which elicit an innate response at the site of infection (1). Neutrophils and macrophages infiltrate the site of tissue infection and release alarmins, including HMGB1 (2). HMGB1 recruits iDCs, resulting in an increase in local mobilization of iDCs (3). Functional DC maturation requires a cross talk with NK cells, which involves HMGB1 expressed at NK-DC synapse (4). Immature to mDC conversion allows DCs to migrate to secondary lymphoid organs (5) and contributes to the enhanced uptake, processing and presentation of microbial antigens to naïve T cells, thus polarizing a Th1 response (6) . This T-cell response involves IL-12 and IL-18 released by mDCs. Cognate interaction between NK cells and DCs may also lead to the selective killing of DCs that are not appropriate for antigen presentation to T cells (7). This editing process, which involves TRAIL, allows NK cells to control the quality of DCs, thus regulating adaptive immunity different strategies to evade DC antiviral activity and to propagate and persist in DCs. HIV-1 replication in DCs is usually weakly productive, and the frequency of HIV-1infected DCs in vivo is much lower than that of HIV-1-infected CD4 þ T cells. 54 However, DCs do not need to be productively infected to transmit the virus to CD4 T cells and to spread it in an infectious form. Analysis of conjugates between DCs and T cells revealed the recruitment of HIV and its receptors CD4, CCR5 and CXCR4 to DC-T cell junction, thus facilitating transmission of HIV during the formation of an infectious synapse in the absence of antigen-specific signaling. 55 This process, known as 'trans-infection', takes place when part of the virus evades classical degradation pathways, being maintained in endosomal acidic compartments, thus retaining viral infectivity for long periods and promoting efficient HIV transfer to CD4 T cells. 56, 57 Trans-infection ability may be restricted to mDCs that display a greater ability to capture incoming virions, retain them in an infectious form in large vesicles within the cells, and transmit them to target CD4 þ T cells, 58, 59 thus augmenting viral dissemination in the lymphoid tissues and significantly contributing to HIV disease progression. Therefore, the viral dissemination that mDCs potentially mediates in vivo is powerful, as viral transmission through trans-infection does not rely on antigen presentation, many CD4 þ T cells being exposed to mDCs-exposing virus. Recently, it has been suggested that HIV can exploit a preexisting exosome trans-dissemination pathway intrinsic to mDCs, thus allowing trans-infection of CD4 þ T cells. 60, 61 Exosomes are membrane vesicles of 30-100 nm in diameter and of endocytic origin, which are produced and secreted in vitro by living cells of diverse origin. They are involved in the stimulation of a specific immune response and they can transfer antigens from infected, tumoral, or antigen-presenting cells to mDCs, increasing the number of DCs bearing a particular antigen, thus amplifying the initiation of primary adaptive immune response. 62 Exosomes from DCs loaded with tumor-derived epitopes on MHC-I molecules are able to stimulate in vivo cytotoxic T lymphocyte-mediated anti-tumor responses, 63 and to indirectly activate in vivo naïve CD4 þ T cells through the exchange of functional peptide-MHC complexes between DCs through a trans-dissemination mechanism. 64 A recent study has shown that upon maturation, DCs are able to capture large amounts of HIV, HIV-gag-VLP, or exosomes, resulting in localization within a CD81 þ compartment, and efficient transmission of captured particles to target T cells in an envelope glycoprotein-independent manner. 65 This 'Trojan exosome pathway' 60 used by mDCs that allows HIV to move between cells in the absence of fusion events could have a prominent role in promoting viral dissemination. DC-dependent activation of NK cells. NK cells are involved in early viral control and by interacting with DCs they have a crucial role of producing pro-inflammatory cytokines and lysing infected cells. In addition, they can interact with T cells and DCs to shape the magnitude and quality of adaptive immune responses. [66] [67] [68] Following tissue invasion by pathogens and subsequent initiation of inflammatory responses, NK cells migrate to lymphoid tissues in response to the chemokines IL-8 and fractalkine (CX3CL1). In a mouse model, Lucas et al. 69 showed that, in vivo, naive NK cells do not acquire effector function unless a priming step has occurred by contact with DCs in draining lymph nodes. The requirements for NK priming was independent of IL-12 and it included secretion of IFN-a by TLR þ cells, which induced upregulation of IL-15Ra and IL-15 expression by DCs, and trans-presentation of IL-15 by DCs to NK cells in the lymph node. NK cells are not fully activated during priming by DCs, as they do not spontaneously produce IFN-g or mediate cytotoxicity. Full activation of NK cells requires an additional contact with mDCs. This results in mutual activation of previously primed NK cells and DC, 70 and in the release of IFN-g by NK cells, thus contributing to antigen-driven T-cell activation. 71 Blocking of IL-12 abolishes DC-induced IFN-g secretion by NK cells, whereas membrane-bound IL-15 on DCs is essential for NK cell proliferation and survival 72 ( Figure 2 ). NK-dependent maturation of DCs. NK/DC cross talk has an important role in the process of DC maturation. NK cells are involved in the positive selection of mature myeloid DCs that, after migration to secondary lymphoid compartments, induce priming of Th cells. 73, 74 Primed NK cells upregulate their cytotolytic function and they release cytokines such as TNF-a and IFN-g, which in turn promote the maturation program of DCs that have captured the antigen. [75] [76] [77] At this stage, NK cells acquire the capability to kill autologous iDC, an event that is dependent upon a process of NK cell activation involving the NKp30 receptor 78 and the TNFrelated apoptosis-inducing ligand (TRAIL)-DR4 pathway. 22 NK cells would spare those DCs that after antigen uptake express high levels of HLA class I molecules, whereas they would kill those DCs (recruited in inflamed tissues) that failed to undergo a full maturation. This process involves inhibitory CD94/NKG2A receptors specific for the non-classical HLA-E molecules. 74 mDCs are resistant to NK killing due to the upregulation of HLA class I expression including HLA-E. 79 mDCs also upregulate an array of additional surface molecules including CCR7, CD80, CD86 and HLA class II, which allow their migration to lymph nodes and their optimal interaction with T cells during their priming phase. Thus, during the early phases of inflammation, DC maturation is under the control of NK cells that have a major role in keeping in check the quality of DC undergoing maturation (Figure 2 ). HMGB1 at the crossroad between innate and adaptive immunity. HMGB1 is expressed at the synapse between NK cells and DCs (Figure 3) , and recent studies highlighted the pivotal role of this cytokine during NK-DC cross talk. Semino et al. 80 showed that NK cells trigger immature DCs to polarize and secrete IL-18 at NK-DC synaptic cleft, thus instructing NK cells to release HMGB1, which promotes DC maturation and protects DCs from lysis. Interestingly, the ability of different NK cell subsets to induce DC maturation is unlinked to their phenotypic and cytolytic features but correlates with the relocation of HMGB1 from the nucleus to the cytoplasm, which is strongly enhanced by engagement of the surface molecule NKp30. 81 Moreover, in the presence of DC-derived cytokines, such as IL-12, a cooperation between NKp30 and DNAM-1 to induce NK cells to kill DCs, release TNF-a and promote DC maturation were evidenced. 82 HMGB1 shows a nuclear localization in primary resting NK cells sorted from the blood, and the NK cell activation induces HMGB1 relocalization from the nucleus to the cytosol followed by extracellular release. 21 During the cross talk between activated NK (aNK) cells and iDC, both NK cells and DCs express HMGB1 21 that appears essential for the upregulation of CD80, CD83 and CD86 maturation markers on DCs and for IL-12 production. The HMGB1 secreted during NK-DC cross talk is also essential for Th1 polarization of naïve CD4 T cells, and RAGE is required for HMGB1 effects on DCs. 43, 21 Moreover, the autocrine/paracrine release of HMGB1 is required for the upregulation on DCs of the CCR7 and CXCR4 chemokine receptors and their migration in response to the chemokines receptor ligands CCL19 and CXCL12, respectively, 40 and RAGE has a nonredundant role in DC homing to lymph nodes, as shown in mice by noninvasive imaging by magnetic resonance. 41 Overall, HMGB1-RAGE pathway is activated in DCs that are committed to maturation in peripheral tissues, and it controls the expression of chemokine receptors in DCs that acquire the ability to reach secondary lymphoid organs where they initiate the clonal expansion of Ag-specific T cells. The disruption of HMGB1-RAGE pathway by specific inhibitors 43, 21 or the genetic deletion of RAGE 41 interrupts this circuit, possibly limiting the initiation of T-cell adaptive response. . NK cells and DCs were stained with red and green Cell Trackers, respectively. During the coculture, one NK cell interacted several times with the DC (pointed out with a star), leading to the killing of the DC. The DC died by apoptosis, as shown by the blebs (indicated with the yellow arrows). This editing process occurred very rapidly, within less than 1 min following the kiss of death by NK cells. 22 (e) Mitochondria rearrangement at NK-DC synapse, detected with a green MitoTracker cells. This inhibitory effect of HMGB1 was caused by repression of LTR-mediated transcription. 83 Extracellular HMGB1 also showed a dichotomic effect in different cell types. Addition of HMGB1 to primary monocytes with active HIV-1 infection was reported to suppress viral replication, associated 19 or not associated 20 to HMGB1-mediated increased release of b-chemokines (RANTES, MIP-1a and MIP-1b), strong inhibitors of HIV entry. In contrast, extracellular HMGB1 increased HIV-1 replication in the chronically infected monocytic cell line U1, 19 a process that did not require de novo protein synthesis. 84 HIV-1 induction relied on HMGB1-RAGE interaction, involved p38, ERK and NF-kB pathway, and stimulated the release of TNF-a. 84 Interestingly, HMGB1 could reactivate ex-vivo quiescent HIV-1 from latently infected PBMC collected in aviremic HIV-infected patients. 84 Thus, HMGB1 may reduce viral replication in acute infection by inducing inhibitors of viral entry, but it may trigger viral replication in latently infected cells, including in cells from HIV-infected patients. NK-DC cross talk contributes to HIV replication through HMGB1. The mechanisms involved in NK-DC interaction during viral infections are poorly understood. It was recently reported in murine CMV (MCMV) infection that MCMVinfected DCs were capable of activating syngeneic NK cells in vitro and also capable of enhancing NK-dependent clearance in vivo, 85 demonstrating the crucial role of NK-DC cross talk in controlling viral replication. In HIV infection, NK-DC interaction was found defective in viremic HIV-1-infected patients, characterized by abnormalities in the process of reciprocal NK-DC activation and maturation. 86 Recently, we investigated the impact of HIV-1 on NKdependent maturation and function of iDCs in an ex-vivo model of cross talk between purified primary NK cells and monocytes-derived iDCs. We discovered that maturation of HIV-1-infected DCs required aNK to occur and this process involved HMGB1. Blocking HMGB1 with specific antibodies or glycyrrhizin, a specific inhibitor of HMGB1, impaired maturation of infected DCs. However, the cross talk between HIV-1-infected DCs and aNK cells was functionally defective, as demonstrated by the strong impairment of DCs to induce Th1 polarization of naïve CD4 T cells. This was associated with the defective production of IL-12 and IL-18 by infected DCs, known to trigger the adaptive response. 21 Moreover, the interaction between aNK and HIV-1-infected DCs resulted in a dramatic increase in viral replication and proviral DNA expression in DCs. This process was mainly triggered by HMGB1, released both by NK cells and DCs, and blocking HMGB1 strongly inhibited HIV replication in both isolated infected DCs and DCs cocultured with aNK cells 21 (Figure 4) . Thus, these findings provide evidence for the crucial role of NK-DC cross talk in promoting viral dissemination, and challenge the question of the in vivo involvement of HMGB1 in the triggering of HIV-1 replication and replenishment of viral reservoirs in AIDS. NK-DC cross talk and HMGB-dependent HIV persistence in DCs. The NK cell-mediated editing process of DCs, which is required to keep in check the quality of DCs prone to mature and further present the antigen to T cells, is compromised during HIV infection. Indeed, NK cells from viremic patients show a decreased ability to kill immature DCs. 87 The defect is associated with an increase in the proportion of CD56 À NK cells with impaired NKp30 function. 86 In addition, increased production of IL-10 during HIV-1 infection can protect immature DCs from NK cellmediated lysis, resulting in accumulation of partially mature, poorly immunogenic DCs in the lymph nodes of infected individuals. 88 In an ex-vivo model of NK-DC cross talk, we showed that HIV-1-infected DCs become resistant to NK cellmediated lysis due to an upregulation in DCs of two apoptosis inhibitors, cellular FLICE-inhibitory protein (cFLIP) and cellular inhibitor of apoptosis protein 2 (c-IAP2). 22 The expression of these inhibitors was upregulated by HMGB1, released by aNK cells at NK-DC synapse, and they protected HIV-1-infected DCs from TRAIL-dependent apoptosis. 22 Blocking HMGB1 with specific antibodies restored the susceptibility of infected DCs to NK killing, and similar effect was observed knocking down c-FLIP or c-IAP2 by siRNA 22 (Figure 5 ). Overall, these findings suggest that impaired NK-DC cross talk during HIV-1 infection is a consequence of bidirectional alteration of both DC and NK cell functions, and they reveal the pivotal role of HMGB1 in HIV-1 persistence in DCs. Extracellular HMGB1 is an important component contributing to tissue injury in acute and chronic inflammatory conditions. HMGB1 and its receptors RAGE, TLR2 and TLR4 have been implicated in mechanisms of many diseases, including cancer, sepsis, atherosclerosis, stroke, rheumatoid arthritis and many other inflammatory conditions. 89 Plasma levels of HMGB1 are elevated during the course of HIV-1 infection 17 and positively associated with high viral load. 90 HMGB1 can be passively released by virus-infected cells including primary CD4 T cells infected with HIV-1, and this was associated with both necrotic and apoptotic cell death. 91 HMGB1 can also be released by non-infected apoptotic CD4 T cells that die through a bystander killing process, which is mainly induced by extracellular HIV-1-encoded proteins and by HIV-1associated chronic immune activation. 46 Increased circulating HMGB1 levels detected in progressive HIV-1 infection, combined with microbial products and TLR ligands, may contribute to gut inflammation and subsequent microbial translocation, suggested to have an important role in HIV pathogenesis. 92 Microbial translocation is the leaking of normally friendly commensal bacteria from the gut -where they are usually contained -into the systemic circulation. Brenchley et al. 92 proposed that this phenomenon contributes to immune activation in patients with HIV, and thus has a causative role in the progression of the disease. Markers of microbial translocation found in the bloodstream include LPS and bacterial DNA, and the level of circulating LPS in the first year of chronic HIV infection is a strong predictor of disease progression independent of CD4 T-cell count and HIV viraemia. 93 The possible link between circulating LPS and HMGB1 levels in inflammatory conditions is suggested by following recent observations: (1) priming of macrophages with LPS induces the processing and releasing of HMGB1 in addition to IL-1b, IL-18 and TNF-a, which requires the inflammasome components ASC, caspase 1 and NALP3; 28 (2) conversely, NALP3 silencing has a protective effect in a murine model of liver ischemia-reperfusion injury, associated with decreased production of HMGB1, IL-1b, IL-18, TNF-a and IL-6; 94 (3) the resistance of caspase-1-deficient mice to LPS correlates with reduced serum HMGB1 levels. 28 (4) HMGB1 forms highly inflammatory complexes with LPS, and signals through the TLR4; 95 (5) an association of elevated circulating levels of LPS and high viral load was reported in HIV-infected patients. 90 Thus, HMGB1-LPS complexes may be important in perpetuating inflammatory amplification loops in HIV disease. Thus, HMGB1 by itself, or combined to LPS or other TLR ligand or cytokines, may induce a self-perpetuating cycle by contributing to immune activation that creates new T-cell targets for viral infection and subsequent increased rate of cell death and release of HMGB1, but also by stimulating HIV replication and viral persistence in DCs. This vicious circle may facilitate HIV disease progression and contribute to AIDS pathogenesis ( Figure 6 ). HMGB1 is an endogenous danger signal that can be released into the extracellular milieu during states of cellular stress or damage, and that is also actively produced by innate effectors such as NK cells subsequently involved in promoting adaptive immunity. HMGB1 has a pivotal role during NK-DC cross talk but, in the context of a chronic viral infection such as that induced by HIV-1, it triggers viral replication in DCs and blocks NK-mediated killing of infected DCs, thus contributing to viral persistence. Increased circulating HMGB1 levels are detected in progressive HIV-1-infected individuals and, in combination with microbial products and TLR ligands, it may also contribute to gut inflammation and subsequently increase microbial translocation and chronic immune activation, a hallmark of HIV-1 disease progression. In order to get a better understanding on the role of HMGB1 in HIV disease and to develop interventions aimed at enhancing immunity to HIV-1, several major questions need to be addressed. The authors declare no conflict of interest. Figure 6 Proposed contribution of HMGB1 to HIV persistence and dissemination. HIV disease progression is characterized by gut inflammation and consequently microbial translocation leading to the release of LPS in the bloodstream (1). Circulating LPS may induce active release of HMGB1 by innate cells, including macrophages and DCs (2) . Necrotic and apoptotic cells that accumulate during chronic HIV infection may also be a constant source of HMGB1 (3). HMGB1 activates DCs by signaling through the receptor RAGE or TLR4 if cooperate with TLR4 ligand LPS, thus triggering NF-kB activation (4) . This results in the release of HMGB1 (5) and proinflammatory cytokines (6) , and the triggering of HIV replication in mDCs (7). Trans-infection of HIV from mDCs to CD4 T cells involves HIV capture by DC-SIGN followed by its recruitment at the site of T-cell interaction (8) . This infectious synapse will lead to the productive infection of CD4 T cells (9) . Trans-infection of HIV can also be mediated by exocytosis of the HIV-1 particles captured by DCs. After endocytosis, the captured HIV-1 particles are targeted to a multi-vesicular endosomal body (MVB) in DCs (10) . Although some of the MVB-localized virus fraction is targeted to the lysosome and degraded to be further presented to TCR in the context of MHC molecules, fusion of MVB with the plasma membrane results in the release of virus particles along with exosomes (11) . Virus produced by infected CD4 T cells, DCs and macrophages spread the infection to the draining lymph nodes and other lymphoid tissue

Evaluation of the Seeplex® Meningitis ACE Detection Kit for the Detection of 12 Common Bacterial and Viral Pathogens of Acute Meningitis

BACKGROUND: Bacterial meningitis is an infectious disease with high rates of mortality and high frequency of severe sequelae. Early identification of causative bacterial and viral pathogens is important for prompt and proper treatment of meningitis and for prevention of life-threatening clinical outcomes. In the present study, we evaluated the value of the Seeplex Meningitis ACE Detection kit (Seegene Inc., Korea), a newly developed multiplex PCR kit employing dual priming oligonucleotide methods, for diagnosing acute meningitis. METHODS: Analytical sensitivity of the kit was studied using reference strains for each pathogen targeted by the kit, while it's analytical specificity was studied using the human genome DNA and 58 clinically well-identified reference strains. For clinical validation experiment, we used 27 control cerebrospinal fluid (CSF) samples and 78 clinical CSF samples collected from patients at the time of diagnosis of acute meningitis. RESULTS: The lower detection limits ranged from 10(1) copies/µL to 5×10(1) copies/µL for the 12 viral and bacterial pathogens targeted. No cross-reaction was observed. In the validation study, high detection rate of 56.4% was obtained. None of the control samples tested positive, i.e., false-positive results were absent. CONCLUSIONS: The Seeplex Meningitis ACE Detection kit showed high sensitivity, specificity, and detection rate for the identification of pathogens in clinical CSF samples. This kit may be useful for rapid identification of important acute meningitis-causing pathogens.

Bacterial meningitis is an infectious disease with high rates of mortality and severe sequelae. Early identification of causative bacterial and viral pathogens is important to enable prompt and appropriate treatment and thereby prevent life-threatening clinical outcomes. Patients with suspected meningitis are generally prescribed antibiotics before hospital admission or cerebrospinal fluid (CSF) collection for examination. However, the admin-istration of antibiotics may render it difficult to culture the causative bacteria, thereby impairing microbiological diagnosis [1, 2] . To counter this problem, alternative methods of molecular identification, including 16S/23S rRNA gene amplification followed by sequencing and real-time PCR, have been developed. However, the clinical application of such techniques is limited because of the presence of interfering bacterial or viral DNA, timeconsuming nature of these assays, and small number of detection channels available on the real-time PCR platform [3] [4] [5] [6] [7] . http://dx.doi.org/10.3343/alm.2012.32. 1.44 www.annlabmed.org These limitations can be overcome by the use of multiplex PCR methods, which enable the rapid and accurate identification of up to 9 bacterial and viral pathogens in clinical samples [8] [9] [10] . We evaluated the diagnostic value of the newly developed Seeplex Meningitis ACE Detection kit (Seegene Inc., Seoul, Korea), a multiplex PCR kit that uses dual priming oligonucleotide (DPO) methods. This kit detects the 12 most common bacterial and viral pathogens of acute meningitis, namely, Streptococcus pneumoniae (SP), Haemophilus influenzae (HI), Neisseria meningitidis (NM), Group B streptococci (GBS), Listeria monocytogenes (LM), herpes simplex virus (HSV)-1 and HSV-2, Varicella-zoster virus (VZV), Epstein-Barr virus (EBV), cytomegalovirus (CMV), human herpes virus 6 (HHV-6), and human enterovirus (HEV). The Seeplex Meningitis ACE Detection kit has 3 components: Seeplex Meningitis-B, which detects 5 bacteria (SP, HI type b, NM, GBS, and LM); Seeplex Meningitis-V1, which detects 6 viruses (HSV-1, HSV-2, VZV, EBV, CMV, and HHV-6); and Seeplex Meningitis-V2, which detects HEV. The Seeplex Meningitis-B and V1 assays amplify DNA, whereas the Seeplex Meningitis-V2 assay amplifies cDNA reverse-transcribed from viral RNA. The target genes amplified from the 12 strains are shown in Table 1 . Each PCR amplification was performed using 5 μL of isolated nucleic acid solution, 2 μL of 10 × primer mixture, and 10 μL of 2 × Multiplex Master Mix (Seegene Inc.) in a total volume of 20 μL. The amplification protocol was as follows: initial denaturation at 94°C for 15 min, 40 cycles of denaturation at 94°C for 30 sec, annealing at 63°C for 90 sec, and extension at 72°C for 90 sec. The amplified PCR products were electrophoresed in 2% (w/v) agarose gels and stained with ethidium bromide. PCR products were cloned into the vector pUC19 (size, 2,686 bp). The cloned plasmids were harvested, DNA concentrations were measured, and the numbers of target copies were calculated (1 µg of a 1,000-bp DNA unit equals 9.1 × 10 11 copies). The sensitivity of the system was evaluated by amplifying 10-fold serial dilutions of each plasmid (concentrations, 10 4 to 10 -1 copies/20 μL), with distilled water as the negative control. Samples were PCR-amplified and electrophoresed in 2% (w/v) agarose gels. Analytical sensitivity was defined as the lowest template copy number for which amplified products were consistently [11, 12] . Bacterial meningitis was differentiated from viral or aseptic meningitis by a WBC count of more than 100 cells/mm 3 with neutrophil dominance and a ratio of CSF glucose/serum glucose level of less than 0.4. Cases with insignificant WBC count or discrepant results in different parameters were classified into the undetermined group. The median patient age was 27 yr (range, 1 month to 72 yr). For the verification study, we tested only the 78 samples collected at the time of diagnosis, i.e., before antibiotic therapy was initiated. Another set of 118 samples was used to assess discrepancies in positive results obtained from the same patients. Samples that were blood-colored, showed leakage, were aged over 12 hr, and strongly suspected of being contaminated (with discrepancies in the pathogens detected in consecutive tests), were excluded. All samples were assessed by conventional CSF microscopic analysis, staining, and culture. Control CSF samples were collected from 27 patients; of these Positive results were confirmed by conventional PCR amplification by using single primer pairs and subsequent sequencing. Amplicons were purified by using a PCR purification kit (Sol-Gent, Daejeon, Korea) and sequenced by using the BigDye Terminator Cycle Sequencing Kit (PE Applied Biosystems, Foster City, CA, USA) and an ABI PRISM 3730XL DNA analyzer (PE Applied Biosystems). Sequences were analyzed using the Basic Local Alignment Search Tool (BLAST) provided by The National Centre for Biotechnology Information (http://www.ncbi.nlm.nih. gov/BLAST). The detection sensitivity was the highest for GBS [lower limit of detection (LoD) of 5 genomes/mL] and the lowest for HI (LoD of 57 genomes/mL). The LoDs for HEV, EBV, VZV, HSV-2, HHV-6, NM, CMV, SP, HSV-1, and LM were 10, 11, 11, 12, 12, 13, 16, 19, 38, and 51 genomes/mL, respectively. The analytical sensi-tivities for all the 12 target pathogens were higher than the 100 copies per reaction stated by the manufacturer. No cross-reaction was evident with human genomic DNA and other reference strains. The Seeplex Meningitis-V2 ACE Detection kit successfully differentiated all the 19 HEV species from the other strains. CSF samples were obtained from 78 patients at the time of diagnosis of active meningitis and before administration of antibiotic treatment. Among these samples, 44 (56.4%) yielded positive results with the Seeplex kit (Table 3 ). Five cases (6.4%) showed positive results for HEV (4) and HSV-1 (1), which were consistent with the results of conventional studies. However, in 2 cases the results were not consistent with the results of conventional studies. One was positive for both LM and EBV in Seeplex Meningitis ACE detection kit, but positive only for LM in the conventional study; the other tested positive for HEV in Seeplex Meningitis ACE detection kit but positive for VZV in the conventional study. Each case presented a bacterial and a viral features, respectively in CSF analysis. The Seeplex Meningitis ACE detection kit detected causative pathogens for 37 cases (47.4%) that tested negative in the conventional studies. Two patients had single bacterial infection with GBS and LM, and CSF analysis revealed bacterial and undetermined features, respectively. Single viral infections were evident in 16 patients with HEV, 9 with VZV, 3 with HSV-1, and 1 each with HSV-2 and EBV. All the cases with positive results for viruses showed viral or undetermined features in CSF analysis. Five patients showed simultaneous infection with 2 bacterial or viral pathogens: 2 tested positive for both SP and HEV and 1 each tested positive for SP and HSV-1, SP and VZV, and EBV and HSV-1. One of the cases positive for both SP and HEV had CSF findings compatible with bacterial meningitis, while the others had features consistent of viral meningitis, with lymphocyte dominance and high glucose level. Sequencing of amplicons showed no evidence of false-positive cross-reactivity; the rates of BLAST match were 98-100%. CSF samples obtained from the remaining 34 patients did not test positive with the Seeplex kit. Sequential work-up with conventional studies showed that 1 patient each was infected with EBV, VZV, SP, HEV, and HSV-1 and had consistent CSF findings. The pathogenic agents in the other 29 patients could not be identified, although CSF analysis indicated that 6 had bacterial infection, 13 had viral infection, and the remaining 10 had indeterminate findings. None of the control cases showed false-positive results. Acute meningitis requires prompt treatment to avoid life-threatening clinical outcomes and severe sequelae. Since various viral and bacterial pathogens can cause acute meningitis, the causative agent must be identified before treatment is initiated. Conventional diagnostic tools, including CSF staining, culture, and antigen analysis, have limited diagnostic sensitivity and specificity. This has prompted the development of several molecular methods employing conventional, real-time, or multiplex PCR. Although previously developed multiplex PCR methods can rapidly identify up to 9 bacterial and viral agents, the Seeplex Meningitis ACE detection kit, which employs a DPO method, can identify the 12 most common bacterial and viral agents that cause meningitis: SP, HI, NM, GBS, LM, CMV, HEV, EBV, HSV-1, HSV-2, HHV-6, and VZV. DPOs, which consist of 2 separate priming regions joined by a polydeoxyinosine linker, yield 2 primer segments with distinct annealing properties. The long 5′-segment initiates stable priming, whereas the short 3′-segment controls target-specific extension; the use of these 2 primer segments effectively eliminates non-specific priming and yields consistently high PCR specificity even under less-than-optimal PCR conditions [13] . This method is highly appropriate for the identification and differentiation of viral and bacterial pathogens with very variable genetic characteristics and low availability of primer sites. Several multiplex PCR kits using the DPO technology have been developed to identify common viral and bacterial pathogens causing respiratory infections and diarrhea, and such kits are currently being used in clinical laboratories [14] [15] [16] . We found that the Seeplex Meningitis ACE Detection kit offered high sensitivity and specificity for identifying bacterial and viral pathogens in patients diagnosed with acute meningitis. The LoDs for the 12 pathogens ranged from 10-50 copies/µL, and no cross-reaction with human genomic DNA or reference (clinically identified) bacterial and viral pathogens was evident, indicating the high specificity of the kit. Our validation study revealed that the kit successfully identified pathogens in the CSF samples collected from 44 of the 78 (56.4%) patients who were clinically diagnosed with acute meningitis. Only 6 of the 44 pathogens identified using the Seeplex Meningitis ACE Detection kit were also identified by other laboratory methods, with the pathogen being HEV in 4 cases and HSV-1 and LM in the remaining 2 cases. Six CSF samples were simultaneously positive for 2 pathogens, suggesting co-infection or contamination rather than false-positivity, because subsequent sequencing of PCR products yielded 98-100% matches with sequences in the BLAST database. Of the 12 pathogens assayed, HEV had the highest prevalence and was detected in 23 samples by the Seeplex Meningitis ACE Detection kit. Of the bacterial strains, SP was the most frequently detected; SP was detected in 5 samples, although it was detected along with viruses, including HEV, HSV-1, and VZV, in 4 samples. Five of the 34 patients, whose samples were negative at the time of diagnosis were later found to have EBV, VZV, SP, HEV, and HSV-1 infection when investigated by conventional studies. These discrepancies were perhaps attributable to inter-specimen differences at varying stages of infection. Of the other 29 patients who tested negative for all pathogens, 10 had indeterminate CSF parameters, whereas 19 showed findings consistent with bacterial or viral infection. The presence of clinical features of meningitis and indeterminate CSF findings in the 10 patients may be attributed to conditions other than acute viral or bacterial meningitis. In contrast, meningitis in the 19 patients with CSF findings consistent with bacterial or viral infections may have been caused by pathogens other than the 12 assayed by the Seeplex Meningitis ACE Detection kit. Alternatively, the concentration of pathogens in the CSF may have been too low to permit detection. In conclusion, the Seeplex Meningitis ACE Detection kit showed high sensitivity and specificity for the 12 most common bacterial and viral agents causing acute meningitis. A high detection rate was observed in the CSF samples obtained from patients clinically diagnosed with acute meningitis. This kit may enable the rapid identification of pathogens in patients with acute meningitis.

ColorPhylo: A Color Code to Accurately Display Taxonomic Classifications

Color may be very useful to visualise complex data. As far as taxonomy is concerned, color may help observing various species’ characteristics in correlation with classification. However, choosing the number of subclasses to display is often a complex task: on the one hand, assigning a limited number of colors to taxa of interest hides the structure imbedded in the subtrees of the taxonomy; on the other hand, differentiating a high number of taxa by giving them specific colors, without considering the underlying taxonomy, may lead to unreadable results since relationships between displayed taxa would not be supported by the color code. In the present paper, an automatic color coding scheme is proposed to visualise the levels of taxonomic relationships displayed as overlay on any kind of data plot. To achieve this goal, a dimensionality reduction method allows displaying taxonomic “distances” onto a Euclidean two-dimensional space. The resulting map is projected onto a 2D color space (the Hue, Saturation, Brightness colorimetric space with brightness set to 1). Proximity in the taxonomic classification corresponds to proximity on the map and is therefore materialised by color proximity. As a result, each species is related to a color code showing its position in the taxonomic tree. The so called ColorPhylo displays taxonomic relationships intuitively and can be combined with any biological result. A Matlab version of ColorPhylo is available at http://sy.lespi.free.fr/ColorPhylo-homepage.html. Meanwhile, an ad-hoc distance in case of taxonomy with unknown edge lengths is proposed.

Many datasets are "naturally" structured as hierarchical classifications. In particular, the Darwin's evolution theory ensures that the relationships between species can be expressed within a tree (named "phylogenic tree"/"taxonomic tree"). However, the exploration of large trees (and graphs) is not easy. 1 Visualising taxonomy together with other pieces of information (obtained from various biological analyses for example) is often extremely instructive; some examples can be found in. [2] [3] [4] In those cases, a level of granularity of the tree is usually chosen and a color is assigned to each subclass thus defined. 4 Several drawbacks can be easily identified. The granularity level is obviously subject to arbitrariness. Moreover, proximity relationships between subclasses are ignored as well as their subdivisions. In the following, we set up an automatic coloring method in order to address these problems. Indeed, we describe a simple method that automatically generates an intuitive color code showing proximity relationships between data in any hierarchical classification. The presented algorithm, named ColorPhylo, associates a specific color to each item so that the taxonomic relationships are shown by color proximity (the closer two items in the tree, the more similar their colors). Colors can thereafter be used in any user's analyses and figures so as to display taxonomy. The above research field is yet relatively unexplored. An automatic coloring tool for tree exploration has been proposed by Fua and co-authors: 5, 6 in order to explore a given taxonomy, leaves are considered as an ordered list of items (the order can result for example from the reading direction when the tree is displayed on a phylogram). Each node can then be related to a color according to a chosen LUT (Look-Up Table) . The LUT may be locally stretched or compressed in order to focus on a given area. Such an interactive control of the LUT allows exploring the taxonomy while considering various depths. This approach suffers however from a major shortcoming: the leaf ordering is not unique (trees are invariant with respect to permutation of linked branches). Very different colorings can then result from various choices. Moreover, close colors can correspond to items that belong to different branches and somewhat different colors can rely to close species. These drawbacks can be observed in Figures 6 and 8 . In the biological field, several tools may be used to color species in taxonomic tree. 4, 7, 8 In particular, PhyloView 4 has been specially designed to display taxonomy and other analyses (phylogenetic trees) together. Strictly speaking, however, these tools are not automatic coloring tools: the user has to define the colors for taxonomic groups that are relevant from his point of view. Node coloring is of main concern in Self-Organising Map (SOM) framework. 9,10 SOM is a non-linear dimensionality reduction method. It is designed to map items (generally from a high dimensional Euclidean space) onto a discrete grid: data are aggregated on each neuron (vector quantisation). Because the grid is discrete, the visualisation of gaps between clusters-if they exist-is uneasy. Various coloring methods have been proposed to account for this drawback, including. [11] [12] [13] [14] [15] [16] [17] In particular, Kaski et al 14, 15 display neurons in a CIELab colorimetric space; 16 Johan Himberg 17 uses a hierarchical classification in order to find clusters and use a cut-off function in the resulting tree so as to linking neurons to a LUT (thanks to a method close to the Fuas' one). 5, 6 However, the SOM context differs from the one of our present objective, despite approaches that are connected. Indeed, we focus here on taxonomies and SOM deals with multidimensional data. Moreover, if hierarchical classifications are often involved in automatic coloration of SOM, the tree is not the input of the algorithm, but a step in the method (especially in the context of cluster discovering). Such methods can be considered, however, as sources of inspiration for the present paper. One of the examples used in the present paper (see section 3.2) is based on the Genbank taxonomy (http://www. ncbi.nlm.nih.gov/sites/entrez?db=taxonomy). Please note that the Genbank taxonomy can not be considered as a conclusive phylogeny, as clearly stated by Genbank owners (http://www.ncbi.nlm.nih.gov/Taxonomy/ taxonomyhome.html/index.cgi?chapter=howcite). We are only looking here for a way to display any given hierarchical classification, whether it is considered as a phylogeny, a taxonomy or so. ColorPhylo aims at assigning a unique color to each species of a given set so that the color differences reflect the taxonomic "distances" between species (see Fig. 1 ): Step 1: If not available, taxonomic distances are calculated from the taxonomic tree. Two procedures are considered depending on whether the edge lengths are known or not (see section 2.2). Thus, a distance matrix between all species is obtained. Step 2: Species are mapped onto a 2D space while preserving the distance matrix as much as possible (section 2.3). Step 3: The map is rescaled (and possibly rotated) (section 2.4) in order to fit a 2D colorimetric subspace (section 2.5). Step 4: Each species located in the colorimetric subspace is consequently given a unique color. The Relation to the colorimetric space each species gets its own color … where the color-code is used. species coloring can therefore be used in subsequent analyses to visualise the taxonomic relationships (as done in section 3). The generation of a color-code for hierarchicallyorganised data. The presented methodology relies on the availability of a distance based on a taxonomic tree. Two cases are possible: I. Edge lengths are known. Distances between species can be immediately deduced: the distance between two species corresponds to the sum of the length of edges connecting them. The "bananaquit bird" dataset offers a good illustration of that case (section 3.1). II. Edge lengths are unknown (the GenBank taxonomy dataset: section 3.2). We then have to (arbitrary) choose the length of edges. Many solutions can be considered. One may assign the same length for all edges. However, such a choice is inappropriate when the tree depth depends on the level of refinement along branches. Such a situation is rather common in taxonomy: the Eukaryote classification is actually more refined than that of Bacteria and Achea (Fig. 2 ). If the length of edges is set constant along the tree, humans appear closer to some bacteria (which are microscopic prokaryotes) than to octopus (which is also an animal). The toy example presented in Figure 3 , left insert (a) provides a good illustration of this bias. In fact, as far as the tree of life is concerned, the distance between two species belonging to the same subclass (branch) must always be smaller than the distance between one of these species and any species that does not belong to the subclass. In order to ensure that this property is always true, we propose to attribute a variable length for edges: an important length is given to edges close to the tree root, and the length is successively reduced when the edge gets away from the root. A geometric progression can be implemented here. The length of edges at the tree root is 1. The length of any other edge equals half the length of the parent edge (Fig. 3 , right insert (B)). This arbitrary choice has a critical property: Proposition: The distance between two species belonging to the same subclass is always smaller than the distance between one of these species and any species that does not belong to the subclass. Proof: Let us consider three species A, B and C where A and B belong to a same subclass, contrary to C: species A and B are linked by node N 1 , and species The DD-HDS projection in the two-dimensional output space is computed (code available at http://sy.lespi. free.fr/DD-HDS-homepage.html). A DD-HDS parameter (λ) can be adjusted in order to more or less emphasize smallest distances: 0 (1 respectively) for minimum (maximum respectively) consideration of longer distances. In the following examples, λ is set to 0.5. The output of DD-HDS is the set of coordinates in a 2D space (coordinates It is worth noting that additive distances (distances in a tree) cannot be perfectly preserved onto a 2D space. However, this fact is not a drawback in the present case because we are less concerned by distances preservation than subclasses segmentation. Indeed, even if the distance mapping is not perfect, it still emphasizes the various levels of the classification thanks to DD-HDS that avoids both "false neighbourhoods" (when a small distance in the output space becomes a large distance in the data space) and "tears" (when a large distance in the output space becomes a small distance in the data space). 19 After being mapped, data are translated and rescaled so as to occupy a circle having a radius equal to 1. This step is required in order to ensure that the map can be related to the circle-shaped HSB colorimetric space (see section 2.5). Coordinates are subsequently cen- , , ″ ″ + and its angle is ϕ ρ 0  and ϕ ρ otherwise. Rotation around the centre and symmetry can take place now. Rotation and symmetry does not impact MDS solutions but allow modifying the color map at will while preserving relative color differences between species (for example, it may be useful to make colors close to the ones the user is familiar with). Each position in the mapping can now be related to a color according the HSB (Hue, Saturation, Brightness) colorimetric conic-shaped space (The HSB space is also called HSV space (V for Value) or HSL space C is linked to node N 1 by node N 2 (similarly to the example in Fig. 3) . d(x, y) corresponds to the distance between elements x and y in the tree. According to the properties of geometric progression with a common ratio equal to 1/2 (basically, Thus, the contribution of edges far from the tree root is reduced, which makes it possible to emphasize most important classes. It must be pointed out that such a procedure should not be considered as a way to assess "true" edge lengths but rather as a heuristic method to account for classes and subclasses within the actual automatic coloring procedure. Whatever the case, the next steps starts from the resulting matrix of taxonomic distances (noted d thereafter). Non-Linear Multi-Dimensional Scaling (MDS) is a set of methods designed to show relationships between items. The explored dataset is displayed as points on a low-dimensional output space. Most of the time, the output space is a Euclidean 2D space, which is proposed as a "map" of data. These techniques are particularly used to explore high dimensional data. Indeed, MDS techniques are expected to retrieve the spatial organisation of high dimensional dataset. To achieve this goal, linear, as well as non-linear Multi-Dimensional Scaling methods preserve the distances between data "as much as possible", according to criteria depending on the method. 18 Usually, small distances are emphasized. As far as our data are concerned, MDS methods are consequently expected to display taxonomic relationships (a distance matrix) onto small dimensional spaces. To achieve this goal, we have chosen the DD-HDS algorithm (Data-Driven High Dimensional Scaling), 19 for its efficiency whatever the distribution of input distances. Evolutionary Bioinformatics 2011:7 (L for Lightness)). The HSB space relates each position in a cone to a color: • Hue corresponds to the angle on a circle (the base of the cone) where 0° is pure red. • Saturation corresponds to the distance to the centre of the circle: 0 for 0% of saturation (pure white), 1 for 100% (pure color). • Brightness corresponds to position on the axe of the cone. No brightness (black) at the cone tip to full brightness at the base. Brightness is not used in the present framework (Brightness is set to 1). A position in a circle is characterised by hue and saturation values (Fig. 4) . Species can subsequently be given a color, norm (ρ i ) being related to saturation and angle (ϕ i ) being related to hue. The continuity of color in the HSB space ensures that color proximity is related to taxonomic proximity Figure 7 (section 3.2). Angle corresponds to hue while distance to centre corresponds to saturation. In the automatic coloring of SOM framework, Kaski et al 14, 15 propose to embed data in a CIELab colorimetric space. 16 Properties of this colorimetric space also fit very well with our context: indeed distance between positions in the CIELab colorimetric space is supposed to be linked with the human perception of color difference. However, significant constraints are placed on the mapping due to the particular shape of the CIELab space. In contrast, the mapping on circleshaped HSB sub-space is easier. Most of the times, data can be expected to fill a large part of the space after a simple rescaling procedure. Moreover, the mapping can be easily rotated and returned in the space in order to find the most convenient color code. It is worth noting that a three-dimensional mapping (such as RGB space) may offer in some instances a better fit with the matrix of distance. However, every color would be eligible, and a species color could be close (or similar) to the background color. With the two-dimensional mapping related to HSB colorimetric space, no species can correspond to grey, which can be chosen as background color. ColorPhylo is tested on two biological applications: the first one refers to a published ornithological study, 20 the second one is related to the analysis of genomic signatures (DNA sequences characterised by oligonucleotide frequencies). It has to be remembered that, on the one hand we study specified knowledge (or analysis results) related to a given dataset (here, birds' geographical positionfirst example-and oligonucleotide counts-second example-) and on the other hand, a taxonomy (the tree) of these data is available. In the present paper, we aim to display the taxonomy and the knowledge together. The procedure described in section 2 provides a taxonomic color-code used here to color geographic position of birds or genomic signatures. Application to the work of E. Bellemain and co-authors: linking geographical distribution and phylogenetic data for bananaquit birds Bellemain and co-authors performed a phylogenetic analysis of bananaquit birds (Coereba flaveola) caught in various places in Latin America and Caribbean islands. 20 Geographical positions of catching sites are displayed (in their Fig. 1 ) and the genetic analyses led to two phylogenetic trees (their Figs. 2 and 3) . Relationships between geographical origin of birds and taxonomy are discussed, and conclusions about the evolutionary history are made. Despite that this work has been successfully achieved without the help of any visualisation tool, a coloring based on the taxonomic tree would have considerably helped these authors. In the following, ColorPhylo is used to assign a color to each bird according to its position in the taxonomic Saturation Hue Figure 4 . hSB colorimetric 2D sub-space (Brightness is set to 1). tree (based on combined mitochondrial data, Figure 2 in) . 20 Here, edge lengths (provided by the phylogenetic analysis) are known and used (see section 2.2, subsection I). Birds are displayed as colored dots, positioned on the map on their catching site (Fig. 5) . Dot coloring allows easy observation of the genetic groups: Lesser Antilles and Puerto Rico (green dots), Bahamas and Quintana Roo (orange dots), continent (purple dots), and Greater Antilles except Puerto Rico (red dots) as well as smaller relationships within groups. We performed the analysis on the same dataset, according to the Fua et al 5,6 coloring method (Fig. 6) . It must be pointed out that the interactive local stretching and compression of the LUT-an essential feature of the Fua et al method-cannot be implemented here (obviously impracticable on a printed document!). For that reason, a comparison between ColorPhylo and the Fua et al algorithm from the present figures exclusively would be highly unfair. However, there are noticeable differences between Figures 5 and 6 . A high diversity in the population of birds from Lesser Antilles and Puerto Rico could be inferred from figure 6 right insert, whereas it is not supported in fact by the evolutionary tree (Fig. 6 , left insert). Moreover, the high difference between birds Confrontation of geographical positions of birds and phylogenetic analysis Figure 6 . Analysis of bananaquit birds data, similar as the one presented in Figure 1 and Figure 5 , except that the Fua et al 5, 6 coloring method is used rather than colorPhylo. from Lesser Antilles and birds from Bahamas and Quintana Roo is clearly underestimated. Dot coloring allows easy observation of the genetic groups: Lesser Antilles and Puerto Rico (green dots), Bahamas and Quintana Roo (orange dots), continent (purple dots), and Greater Antilles except Puerto Rico (red dots) as well as smaller relationships within groups. Coloring according to the Genbank tree: Species paths are from Genbank so that "root, cellular organism, Eukaryota, Fungi/Metazoa group, Metazoa; Eumetazoa, Bilateria, Coelomata, Deuterostomia, Chordata, Craniata, Vertebrata, Gnathostomata, Teleostomi, Euteleostomi, Sarcopterygii, Tetrapoda, Amniota, Mammalia, Theria, Eutheria, Euarchontoglires, Primates, Simiiformes, Catarrhini, Hominoidea, Hominidae, Homo/Pan/Gorilla group, and Homo Sapiens" qualifies human for example. Paths are derived from a rooted tree where edge lengths are unknown. Taxonomic distances are calculated according to the procedure described in section 2.2, subsection II. Colors are then selected according to section 2.3, 2.4 and 2.5. Data embedded in the colorimetric space are displayed in Figure 7 . Species are embedded in a two-dimensional space generated from the taxonomic distance matrix. Colors are assigned from the position in the mapping (by construction, colors are smoothly Contrasting with figure 7 , the three domains of life are not clearly segmented by color in Figure 8 . An example of color code use: Study of the link between taxonomic proximity and genomic signature proximity: The whole set of short oligonucleotide frequencies observed in a DNA (Deoxyri-boNucleic Acid) sequence is species-specific and is thus considered as a "genomic signature". 2,21 Moreover, a DNA segment as short as 1 Kb (kilobase) is sufficient to characterise the genomic signature of the species. As a consequence, genomic signature appears qualifying the "writing style" of the species. The genomic signature is species specific: it allows finding the species of origin of a DNA fragment with a fairly good efficiency. 2, 22 Note that because the genomic signature is stable along the genome, non-homolog fragments of species can be compared (homology is required for most other methods devoted to comparative genomics). Lastly, proximity between species in terms of genomic signature is known to be linked to evolutionary proximity. Indeed many phylogenies based on hierarchical classifications of signatures have been proposed. [23] [24] [25] [26] [27] [28] Another powerful approach to describe the taxonomic organisation of genomic signatures uses dimensionality reductions: data are embedded on a two-dimensional (or sometimes three-dimensional) space. These representations have been achieved by Principal Components Analysis (PCA), 2,3 by Self Organising Map (SOM) 29 or by Data-Driven High Dimensional Scaling (DD-HDS). 19 However, details of the spatial organisation of genomic signatures cannot be easily observed on a large dataset because of the limitation in term of discrete colors. In the following example we propose to visualise the taxonomic tree of species on the DD-HDS mapping of genomic signatures by means of the color code provided by ColorPhylo and Figure 7 . Prior to mapping, signatures are corrected using 1-order Markov model 30 as recommended in. 25, 27 A distance matrix between signatures is then obtained, based on the Pearson's correlation coefficient as recommended by. 26 The mapping is subsequently generated: spatial proximity between items expresses proximity between species from the genomic signature point of view. Colors are provided by the ColorPhylo procedure (the map of species in the related taxonomic color-space is shown in Fig. 7) : color proximity between items relies on taxonomic proximity between species. As a result, the various levels of taxonomy are simultaneously observable on a single figure (Fig. 9) . The taxonomic organisation of signatures is clearly demonstrated. In particular, the patches of homogeneous colors support concluding that the similarity between genomic signatures accurately matches the taxonomic tree of the species, the signatures come from. Using ColorPhylo is straightforward. Hierarchical classifications can be easily displayed together with their relationships with any other organisation of the data. We have observed on real life examples that the interpretation of the resulting color code is fully intuitive. In the first example, geographical origins of bananaquit birds are related to phylogenetic data in order to analyse the evolutionary history. However, because the number of items and the complexity of the dataset was somewhat low, Bellemain and co-authors succeeded in describe the relationships between taxonomy and geographical distribution in their publication (but at the price of more irksome work). When the size of the dataset and/or the complexity of their relationships increase, ColorPhylo can provide a critical benefit, as it can be observed for the second application. In that example, (genomic signature) colors express the membership of items to one of the three domains of life, with subtle shades showing subclasses. In both cases, we have instantaneously access to the structure of classification through an attractive visualisation plot. Although variations of color are theoretically unlimited, we rely on the perceptual discriminative power of the human eye. Surprisingly, the method gives access to a remarkable degree of detail (well above what is expected with a "manually" defined color code). In addition, a focus on a small region of the data is always possible by an ad hoc local color reallocation. Similarly, Colorphylo may be adapted to fit color-impaired users' requirements. The 2-dimensional color-space can be modified at will according to any desired effect, including of course satisfaction of the user's color perception. In this paper, we have proposed a method to study the relationship between a given knowledge on a set of data (here, the birds' geographical position and the oligonucleotide frequencies in DNA sequences of species) and a specific organisation of these data, expressed by a taxonomic tree. Our approach can easily be adapted to other contexts. For example, if the organisation of the data results from an analysis based on a distance matrix (such as the ones performed by Neighbor joining, 31 Fitch-Margoliash, 32 …), the original distance matrix may be preferred to the taxonomic distance (such an approach may have been implemented for the bananaquit birds dataset). In fact, the procedure may be extended to the analysis of any kind of organisation of the data, given it is expressible as a distance matrix for which a 2D mapping makes sense. A matlab version of ColorPhylo is available at http://sy.lespi.free.fr/ColorPhylo-homepage.html. SL and BF conceived the method together and wrote the manuscript. Both authors read and approved the final manuscript. We thank Eva Bellemain and collaborators, as well as Biomed Central, for the permission to reproducing their figure. We also thank Mikael Cugnet for its useful comments. Author(s) have provided signed confirmations to the publisher of their compliance with all applicable legal and ethical obligations in respect to declaration of conflicts of interest, funding, authorship and contributorship, and compliance with ethical requirements in respect to treatment of human and animal test subjects. If this article contains identifiable human subject(s) author(s) were required to supply signed patient consent prior to publication. Author(s) have confirmed that the published article is unique and not under consideration nor published by any other publication and that they have consent to reproduce any copyrighted material. The peer reviewers declared no conflicts of interest. In the Bananaquit bird example, a very similar color has been given to birds from all great Antilles Islands. We subsequently run ColorPhylo again while focusing on the Antilles Islands subclass. The result is displayed on Annexe Figure 1 . The tight matching between phylogeny and geographical data is demonstrated in details by the color code. publish with Libertas Academica and every scientist working in your field can read your article "I would like to say that this is the most author-friendly editing process I have experienced in over 150 publications. Thank you most sincerely." "The communication between your staff and me has been terrific. Whenever progress is made with the manuscript, I receive notice. Quite honestly, I've never had such complete communication with a journal." "LA is different, and hopefully represents a kind of scientific publication machinery that removes the hurdles from free flow of scientific thought." Your paper will be: • Available to your entire community free of charge • Fairly and quickly peer reviewed • Yours! You retain copyright http://www.la-press.com

Deconstructing host-pathogen interactions in Drosophila

Many of the cellular mechanisms underlying host responses to pathogens have been well conserved during evolution. As a result, Drosophila can be used to deconstruct many of the key events in host-pathogen interactions by using a wealth of well-developed molecular and genetic tools. In this review, we aim to emphasize the great leverage provided by the suite of genomic and classical genetic approaches available in flies for decoding details of host-pathogen interactions; these findings can then be applied to studies in higher organisms. We first briefly summarize the general strategies by which Drosophila resists and responds to pathogens. We then focus on how recently developed genome-wide RNA interference (RNAi) screens conducted in cells and flies, combined with classical genetic methods, have provided molecular insight into host-pathogen interactions, covering examples of bacteria, fungi and viruses. Finally, we discuss novel strategies for how flies can be used as a tool to examine how specific isolated virulence factors act on an intact host.

Drosophila has emerged as an important model for examining the function of genes that are relevant to diverse human diseases affecting a broad range of cell types (for reviews, see Bier, 2005; Bier and McGinnis, 2008) . Additionally, this model organism can serve as a host for a surprising variety of bacterial and viral pathogens. Seminal discoveries in the field of host-pathogen interactions have been made in Drosophila. For example, the Toll signaling pathway, which plays a central role in innate immunity, was first identified in Drosophila, and studies using this model organism have helped to identify and delineate fundamental conserved host genetic pathways involved in barrier formation and maintenance (Mace et al., 2005; Martin and Parkhurst, 2004; Pearson et al., 2009; Ting et al., 2005) , innate immune signaling (Agaisse and Perrimon, 2004; Dionne and Schneider, 2008; Ferrandon et al., 2007; Igaki et al., 2010; Ryu et al., 2010) , the RNA interference (RNAi) response (Sabin et al., 2010) , pathogen engulfment (Meister, 2004) , and the evolution of intracellular pathogens (Haselkorn, 2010; Serbus et al., 2008) . As discussed in detail below, Drosophila has also been used to identify and analyze the function of pathogen-derived virulence factors (Avet-Rochex et al., 2005; Avet-Rochex et al., 2007; Botham et al., 2008; Guichard et al., 2010; Guichard et al., 2006; Shelly et al., 2009; Guichard et al., 2011) . Many genetic tools are available to address the mechanisms of pathogen action in Drosophila. These include comprehensive genetic screens, or genome-wide RNAi screens, in cell lines and intact flies that can identify host pathways required to defend against pathogens. Reciprocally, it is also possible to search for pathogen-encoded factors that are required for virulence in flies. In vivo studies in flies are greatly facilitated by the ability to direct expression of transgenes encoding host or pathogen proteins in specific cell types using the GAL4-UAS transactivation system (Brand and Perrimon, 1993) or the new independently acting LexA system, which allows for combinatorial expression of genes in distinct or overlapping patterns (Yagi et al., 2010; Pfeiffer et al., 2010) . Moreover, it is possible to perform epistasis experiments using combinations of dominant and recessive mutations (or mutants with opposing phenotypes) in a given pathway to determine the sequence in which genes act in that pathway. These versatile tools, combined with the rapid Drosophila life cycle, allow detailed genetic analysis of virulence factors that act on tissues or organs; such experiments would be much more difficult to conduct in vivo in mammalian systems. In this review, we first outline the basic host defense mechanisms used by Drosophila to resist and respond to invading pathogens to provide context for our discussion of how flies can be used to deconstruct key mechanisms of host-pathogen interactions. We then focus on three related topics: (1) genome-wide RNAi screens in Drosophila cell lines infected with pathogens to identify host pathways for defense or that are exploited by pathogens (e.g. bacteria, fungi, viruses); (2) classical genetic and RNAi screens conducted in intact flies to delineate host defense pathways that are active in specific tissues (e.g. the gut) or to identify important virulence factors produced by the pathogen; and (3) analysis of the function of specific pathogen virulence factors in an intact organism. The studies reviewed here highlight the speed and power of Drosophila genetics for uncovering new pathways and factors in host-pathogen interactions, as well as for characterizing unknown activities of specific virulence factors. Identification of such elements in the host-pathogen relationship should help to guide studies in vertebrate systems and contribute to defining new targets for potential therapeutic intervention. Host-pathogen interactions in Drosophila PERSPECTIVE and an internal component comprising the gut (or endoderm). The formation of both the outer epithelial barrier and the inner intestinal barrier depends on the formation and maintenance of intercellular junctions, and many basic discoveries in this field have been made in Drosophila (Banerjee et al., 2006; Furuse and Tsukita, 2006; Wirtz-Peitz and Zallen, 2009 ). Such studies have delineated key mechanisms involved in establishing apical-basal polarity, including the assembly of distinct protein complexes at adherens junctions and septate junctions (claudin-dependent junctions that share important similarities with vertebrate tight junctions). One highly conserved feature of this process is the role of the exocyst protein complex in trafficking proteins such as cadherins and cell signaling components to adherens junctions (Andrews et al., 2002; Beronja et al., 2005; Blankenship et al., 2007; Jafar-Nejad et al., 2005; Langevin et al., 2005; Mehta et al., 2005; Murthy et al., 2003; Murthy et al., 2005; Murthy and Schwarz, 2004; Murthy et al., 2010) . At first glance, the mammalian epidermis seems very different from that of flies ( Fig. 1A ), but there are striking parallels with respect to the formation and maintenance of epithelial barriers in the two species, illustrating a probable common ancestral origin. For example, claudin-family proteins forming the tight junctions between epithelial cells seem to have similar functions in both flies (Behr et al., 2003; Nelson et al., 2010; Wu et al., 2004) and mice (Furuse et al., 2002) [see the 2009 article by Furuse for a review on the role of claudins and other tight junction proteins in mammalian epithelia (Furuse, 2009) ]. Similarly, the transcription factor Grainyhead (Grh) plays an important role in regulating the expression of genes that are required to form the cross-linked outer epidermal surface both in flies (Bray and Kafatos, 1991) and mice (Matsuki et al., 1998; Ting et al., 2005) (although the set of Grh target genes seems to be different in each species). Grh also regulates genes that are involved in wound repair both in flies (Mace et al., 2005) and mice (Ting et al., 2005) . It is noteworthy that Drosophila and vertebrate intestinal epithelia are also similar in several respects. These parallels include: the fact that stem cells play an important role in replacing cells that have undergone pathogen-dependent apoptosis; the sequential deployment of Wnt and Hedgehog (Hh) signaling during the differentiation of intestinal epithelial cells (Pitsouli and Perrimon, 2008; Takashima et al., 2008) ; and the formation of the morphologically specialized brush border microvilli and the underlying cytoskeletal terminal web (Li et al., 2007; Morgan et al., 1995; Phillips and Thomas, 2006) . A challenge faced by intestinal cells is that they must tolerate commensal bacteria, with which they have a mutualistic relationship (Backhed et al., 2005; Dale and Moran, 2006; Sansonetti and Medzhitov, 2009) , while also mounting a vigorous response to pathogens (for a review, see Ryu et al., 2010) . One important pathway involved in this distinction controls the production of reactive oxygen species (ROS) by the dual-oxidase (Duox) transmembrane protein (Ha et al., 2009; Ha et al., 2005a; Ha et al., 2005b) , which also plays a key role in the human gut (for a review, see Ryu et al., 2010) . Genetic analysis in Drosophila has revealed bi-stable control of Duox activity in the gut. In the presence of commensal bacteria and absence of pathogenic species, low-level activation of the immune deficiency (IMD) pathway of the innate immune system (see later) induces negative feedback of the Duox pathway (at both the level of expression and activity), resulting in low basal levels of ROS production. By contrast, when invading pathogens are detected by host immune signaling, expression and activity of Duox components is greatly increased, leading to destruction of the pathogenic bacteria (Ha et al., 2009) . The inducible ROS-producing Duox system works in parallel with other immune pathways, such as the Jun N-terminal kinases (JNK) pathway. JNK signaling is activated in intestinal epithelial cells of adult flies following ingestion of pathogenic Pseudomonas aeruginosa , which leads to proliferation of intestinal stem cells to compensate for apoptotic loss of mature infected cells (for a review, see Pitsouli et al., 2009 ). An interesting aspect of this pathogen in Drosophila is that, in combination with an activated oncogenic form of RAS, it can lead to overproliferation of stem cells to form tumors . Whether the elevated incidence of human cancers of the intestinal tract as a result of associated bacterial infection (Bornschein et al., 2009; Selgrad et al., 2008) is similarly influenced by RAS activation remains to be determined. Another interesting emerging theme is the elucidation of host pathways involved in detecting cell damage in the intestine, which then regulate stem cell mediated repair of the damaged epithelium. These studies have revealed important contributions of the insulin (Amcheslavsky et al., 2009 ) and TSC-TORC1 (Amcheslavsky et al., 2011) pathways, as well as of Hippo (Hpo)-mediated activation of the JAK-STAT and Epidermal growth factor receptor (EGFR) pathways (Ren et al., 2010) . Finally, experiments involving oral infection of flies with Erwinia carotovora, a natural Drosophila pathogen, suggest that gut homeostasis is maintained by active tissue repair of cell damage caused by bacteria (Buchon et al., 2009) . The observation that ROS can trigger apoptosis followed by repair in the larval gut (Gupta et al., 2010) suggests that the Duox pathway provides compensatory feedback to pathways controlling apoptosis and stem cells to ensure that host cells damaged by ROS exposure are duly replaced. Overall, these studies provide an excellent foundation for further analysis of how the gut responds to pathogens by repairing damage and differentially responding to commensal versus pathogenic bacteria. Broadly speaking, the innate immune response consists of three parts: (1) pathogen immobilization (Fig. 1B) , (2) core immune signaling pathways (Toll, IMD and JAK-STAT) (Fig. 1C) and (3) the RNAi pathway (Fig. 1D ). Because there have been several excellent reviews describing these pathways, we only summarize here their key elements, as depicted in Fig. 1 , and refer the reader to other sources for more in depth descriptions (Agaisse and Perrimon, 2004; Akira et al., 2006; Bhavsar et al., 2007; Brodsky and Medzhitov, 2009; Diacovich and Gorvel, 2010; Dionne and Schneider, 2008; Ferrandon et al., 2007; Folsch et al., 2003; Sansonetti, 2008) . The most basic innate response to bacterial or fungal infection is a cellular response (Jiravanichpaisal et al., 2006) that immobilizes the invading microbe by phagocytosis, engulfment or a Host-pathogen interactions in Drosophila PERSPECTIVE melanization reaction that traps it (Fig. 1B) . Pathogens can also be immobilized in flies and other insects by a clotting reaction (Dushay, 2009) . Once immobilized, the pathogen can then be either destroyed extracellularly by antimicrobial peptides (AMPs) or eliminated intracellularly. Three basic types of Drosophila blood cell (known as hemocytes) perform these functions: plasmatocytes, which are professional phagocytic cells akin to mammalian macrophages; lamellocytes, which wrap themselves around invading microorganisms to form an enveloping capsule; and crystal cells, which contain the enzymes that catalyze melanization (Meister, 2004) (Fig. 1B) . As discussed later in more detail, many host genes that are required for phagocytosis have been identified using Drosophila in a series of genome-scale cell-based screens. Similar studies in the future might shed light on genes that are essential for lamellocyte and crystal cell function. Autophagy is another general mechanism important for clearing bacteria (Yano et al., 2008) and viruses (Cherry, 2009; Shelly et al., 2009) . It should be pointed out, however, that autophagy can also be hijacked for the benefit of the pathogen, as in the case of poliovirus, which derives its envelope membranes from autophagic vesicles (Suhy et al., 2000) . The second part of the Drosophila innate immune response comprises a set of core signaling pathways (Fig. 1C) : the Toll pathway, the IMD pathway and the JAK-STAT pathway. The activities of these pathways are modulated by other pathways, such as that mediated by target of rapamycin (TOR) or Eiger-Wengen [Drosophila homologs of human tumor necrosis factor (TNF) and TNF receptor]. When induced following pathogen infection, innate immune pathways result in the production of AMPs such as Drosomycin and Diptericin (Dionne and Schneider, 2008; Ferrandon et al., 2007; Agaisse and Perrimon, 2004; Dionne and Schneider, 2008; Ferrandon et al., 2007; Folsch et al., 2003) . The third part of the Drosophila innate immune response is the double-stranded RNAi pathway that is involved in defending against many types of viral infections, and which also protects against viral infection in plants and animals (for a review, see Sabin et al., 2010) (Fig. 1D ). The RNAi pathway is activated by viral nucleic acids and can be broken down into two main steps: (1) biogenesis of 21-base-pair double-stranded viral small interfering RNAs (siRNAs), which is accomplished by the Dicer protein complex, and (2) the silencing of viral RNAs by the host-induced viral siRNAs, which is accomplished by the RNA-induced silencing complex (RISC). This innate protective system has been highly amenable to analysis using genome-wide screening in Drosophila cells (see below). One of the great recent technical advances in the field of Drosophila cell biology has been the development of efficient whole genome RNAi screens to identify genes required for specific cellular processes (Mohr et al., 2010; Perrimon and Mathey-Prevot, 2007; Perrimon et al., 2010) . In such assays, Drosophila cell lines such as hemocyte-derived S2 cells or Kc cells (which can be induced by hormone treatment to differentiate into neurons) are grown in 384well plates and treated with a library of double-stranded RNAs that have been designed for highly selective RNAi-mediated knockdown of each of the predicted Drosophila coding messenger RNAs (mRNAs). These cells are then assayed for performance of a cellular process such as cell viability, cell shape changes or bacterial uptake by phagocytosis. By screening such libraries in replicate and then re-screening RNAi candidates that test positive for a specific effect, it is possible to approximate genome-wide coverage of all genes required in these cells for a given process [for an excellent, comprehensive review of such RNAi screens, see Cherry (Cherry, 2008) ]. Such screens have been used to identify many host response factors that are crucial during infection by bacteria, fungi and viruses. Several straightforward RNAi screens have been conducted to identify genes that are required for phagocytosis of various species of bacteria by S2 cells. For these experiments, ingestion of bacteria expressing green fluorescent protein (GFP) is monitored and host genes involved in phagocytosis are revealed on the basis of the identity of specific RNAi molecules that inhibit uptake of fluorescence. These screens have revealed that distinct sets of host genes are essential during infection by various pathogens. For example, different pathogens are recognized by distinct cell surface receptors, such as peptidoglycan recognition proteins (PGRPs) (Ramet et al., 2002) , SR-C1 (Ramet et al., 2001) , Eater (Kocks et al., 2005) , Nimrod (Kurucz et al., 2007) or DSCAM (Watson et al., 2005) . However, these screens also defined a core set of intracellular uptake components that are regulated in all types of bacterial infection tested: these included genes required for actin remodeling (e.g. genes encoding proteins of the Arp2/3 complex) and endocytosis (e.g. COPI and COPII), as well as genes encoding factors that are required to recycle endosomes to the cell surface, such as proteins in the exocyst complex (Agaisse et al., 2005; Cheng et al., 2005; Philips et al., 2005; Ramet et al., 2002; Stroschein-Stevenson et al., 2006; Stuart et al., 2007) . Other genes involved in the response to bacterial infection that have been identified in RNAi screens are required for host cells to clear ingested bacteria. Again, these screens defined a set of generally required genes that limit bacterial survival or replication, such as genes encoding endosomal sorting complex required for transport (ESCRT) proteins (Philips et al., 2008) , as well as genes preventing the growth of specific pathogens, such as lysosomal hexosaminidase, which restricts growth of Mycobacterium marinum but not Listeria monocytogenes or Salmonella typhimurium (Koo et al., 2008) . In other standard genetic studies, intracellular microorganisms such as Wolbachia were found to also engage in mutualistic symbiotic relationships with the host, such as protecting the host against viral infection (Hedges et al., 2008; Teixeira et al., 2008) and nutritional supplementation (Brownlie et al., 2009) , which presumably arose during co-evolution of the endosymbiont and host. RNAi technology can also be used in a combinatorial fashion to knock down the activity of two or more genes at a time, which permits detection of genes acting in parallel in a given process or pathogenic infection. In one study, Dorer and colleagues performed Host-pathogen interactions in Drosophila PERSPECTIVE a series of single and double gene knockdown experiments of 73 genes in Kc cells to test the hypothesis that Legionella pneumophila, the agent of Legionnaires' disease, recruits membrane material from endoplasmic reticulum (ER)-to-Golgi trafficking (Dorer et al., 2006) . Although few single knockdowns had much of an effect, the authors found evidence supporting their hypothesis in several double knockdown experiments. For example, double knockdown of the intermediate compartment and Golgi-tethering factor transport protein particle (TRAPP) together with the ER SNARE protein Sec22 resulted in reduced pathogen replication efficiency. They also showed a requirement in bacterial replication for the Cdc48-p97 complex that is involved in ER-associated degradation, and demonstrated that this complex is also important for Legionella pneumophila replication in mouse bone-marrow-derived macrophages. These studies underscore the role of endocytosis in phagocytic host cells and, owing to the combinatorial power of the system used, revealed a role for endocytic steps carried out by parallel mechanisms. Fungi generally activate the Toll signaling pathway of the Drosophila innate immune system via a specific set of PGRP detection peptides (Fig. 1C ). RNAi screens similar to those performed to identify host genes required for phagocytosis of bacteria have also been carried out to identify host factors involved in response to fungi such as Candida albicans (Stroschein-Stevenson et al., 2006; Stroschein-Stevenson et al., 2009) . Beyond identifying genes with broad expected functions, such as regulators of the actin cytoskeleton and vesicular trafficking, these studies also identified genes required for the uptake of specific fungal pathogens. One of these proteins, Macroglobulin complement related (Mcr), is a secreted protein that binds directly to C. albicans and promotes its internalization. Interestingly, Mcr is related to four other Drosophila thioester proteins (Teps), two of which are selectively required for phagocytosis of specific bacterial species (TepII for Escherichia coli and TepIII for Staphylococcus aureus), but not for phagocytosis of C. albicans (Stroschein-Stevenson et al., 2006) . In addition to being susceptible to infection by bacterial and fungal pathogens, Drosophila is also a natural host for viruses such as Drosophila C virus (DCV), Drosophila X virus (DCX) and Flock House virus, and, perhaps surprisingly, by a broad variety of viruses causing disease in humans such as Sindbis virus, vesicular stomatitis virus (VSV; a virus of the Rhabdoviridae family, which includes the well-known rabies virus), Rift Valley fever virus, dengue virus and West Nile virus (Cherry et al., 2005; Cherry et al., 2006; Cherry and Perrimon, 2004; Galiana-Arnoux et al., 2006; van Rij et al., 2006; Wang et al., 2006) . Genome-wide RNAi screens have identified several important host factors that are exploited by viruses, such as factors required selectively for replication of influenza virus (Hao et al., 2008) or propagation of dengue virus (Sessions et al., 2009) . Similarly, viruses such as DCV that have transcripts with internal ribosome-binding sites depend on several host translation factors that are not required for other types of viruses lacking these sites (Cherry et al., 2005) . DCV also requires the host factor COPI to generate a vesicular compartment, which is necessary for viral replication, and COPI is also required for the replication of the related poliovirus in human cells (Cherry et al., 2006) . As another example, infection by vaccinia virus (the prototypical poxvirus) was found to depend on the AMP-activated kinase (AMPK) complex, the master energy sensor of the cell, for endocytic entry and actin remodeling (Moser et al., 2010) . The authors found a similar requirement for AMPK in facilitating vaccinia infection of mouse embryonic fibroblasts and showed that this kinase was also involved in viral entry via the process of macropinocytosis. As mentioned above, the RNAi pathway plays a key role in defending against viral infection. Genome-wide and targeted RNAi screens have contributed to the elucidation of this pathway (Galiana-Arnoux et al., 2006; Nayak et al., 2010; Otsuka et al., 2007; Sabin et al., 2009; van Rij et al., 2006; Wang et al., 2006) (for a review, see Sabin et al., 2010) and the importance of the systemic spread of an RNAi activating signal (probably some large viral doublestranded RNA) for stimulating RNAi-dependent immunity throughout the organism (Saleh et al., 2009) . Interestingly, siRNAs do not spread from cell to cell in Drosophila (Roignant et al., 2003) , in contrast to the mechanism by which RNAi molecules are directly distributed in plants (Palauqui et al., 1997; Winston et al., 2002) and nematodes (Fire et al., 1998; Voinnet et al., 1998) to mediate systemic immunity. As a complement to cell-based screening methods, it is also possible to screen for host genes that are required to combat pathogen infection using intact flies. Although these screens are more laborious than screens in Drosophila cell lines, or wholegenome RNAi screens in worms (i.e. Caenorhabditis elegans screens can be done on plates), screens using intact flies can be accomplished either by classic mutagenesis or by screening high quality collections of stable UAS-RNAi stocks. A great advantage of the latter approach is that one can use the GAL4-UAS expression system (Brand and Perrimon, 1993) to drive expression of UAS-RNAi constructs throughout the organism or in specific subsets of cells or stages of development (Fig. 2B ). For such experiments, a strain of flies carrying a transgene under the control of the yeast upstream activating sequence (UAS) is crossed to a strain of flies expressing the GAL4 transcription factor (which binds to the UAS sequence and activates transcription in a particular pattern, e.g. in the gut). The progeny then express the UAS transgene of interest in the pattern determined by the GAL4 'driver' stock, permitting expression of genes in specific cell types at specific stages of development. This level of control permits investigators to identify the cells or organs in which gene functions are required [e.g. epidermis, fat body (the main source of systemic AMPs, and an approximate model of the mammalian liver), hemocytes or gut]. In one screen using adult flies, host defense factors that are required to protect against intestinal infection with the opportunistic broad-host-spectrum pathogen Serratia marcescens were first identified by using a large collection of fly lines in which 13,000 individual RNAi molecules were used to knock down target gene expression throughout the organism (Cronin et al., 2009) . RNAi molecules that caused increased lethality following infection were then tested further for their role in defending Host-pathogen interactions in Drosophila PERSPECTIVE against S. marcescens infection of the gut by expressing the relevant UAS-RNAi constructs with gut-specific and hemocytespecific drivers. These studies first confirmed the dependence on the IMD (but not Toll) innate immune pathway for responding to infection by S. marcescens (as would be expected for a Gramnegative bacterium), and also revealed an important role for the JAK-STAT pathway in responding to infection in the gut (Fig. 2C ). Further analysis of JAK-STAT signaling showed that this pathway regulates stem cell proliferation and thereby intestinal epithelial homeostasis during infection. These results obtained in intact flies provide an important complement to screens performed in C. elegans, which also identified several signaling systems important for innate immunity (Irazoqui et al., 2010) . Studies of damage and repair by gut pathogens can be conducted in Drosophila because flies, but not worms, have intestinal stem cells that replenish epithelial cells after they undergo programmed cell death during infection (see above). Systematic screens such as that mentioned above (Cronin et al., 2009) can also be used to identify host factors co-opted by a pathogen that, when mutated, render the host resistant to the pathogen. An example using a traditional genetic approach is the case of Vibrio cholerae, in which investigators showed that feeding V. cholerae bacteria to flies caused rapid death (i.e. in 2-3 days) that required the function of the primary virulence factor cholera toxin (CTX) (Blow et al., 2005) . CTX is an ADP ribosyl transferase that specifically ribosylates the Gs subunit of a host trimeric Gs protein, resulting in constitutive activation of adenylate cyclase (Middlebrook and Dorland, 1984) . The dependence on CTX was Host-pathogen interactions in Drosophila PERSPECTIVE unexpected because flies lack the enzymes required to synthesize the GM1 ganglioside that serves as the CTX receptor and that is present in most vertebrates and a few invertebrates. Accordingly, feeding flies purified CTX holotoxin had no effect (Blow et al., 2005) . Paradoxically, however, full virulence of ctx-mutant bacteria could be restored by feeding infected flies purified CTX, suggesting that, in the presence of the bacteria, a novel alternative route of CTX delivery to host cells in the gut might be employed. Further analysis showed that several host target factors known from mammalian studies were required by V. cholera to infect flies, such as proteins mediating the dehydrating effects of CTXdependent cAMP production -including a Gs subunit, adenylate cyclase and an SK-type potassium ion channel (Blow et al., 2005) . Having established flies as a model for V. cholerae infection, the authors then screened a large collection of stocks with mapped transposon insertions into the fly genome and identified mutations that either enhanced or reduced severity of infection. This strategy identified several host genes important for the response to V. cholerae infection, including those conferring resistance when mutated and that presumably are exploited by bacteria (e.g. components of the TNF and IMD pathways) as well as those used in host defense (e.g. the apoptotic pathway) (Berkey et al., 2009 ). Adult flies and cells have also been used to screen for pathogenencoded factors that contribute to virulence. One particularly elegant screen for bacterial virulence factors was carried out for P. aeruginosa, an opportunistic human pathogen that can cause serious disease. Over 4000 transposon insertion mutants of the bacterium were screened by injecting them into the adult fly hemolymph and measuring percent lethality. This resulted in the identification of 15 different bacterial loci that contributed significantly to virulence (Kim et al., 2008) . The authors examined the basis of virulence for one of these genes, hudR. hudR encodes a transcription factor that represses expression of the neighboring gene hudA, which is involved in ubiquinone biosynthesis. On the basis of their genetic analysis, the authors hypothesized that the decreased virulence of hudR mutants resulted from overexpression of hudA. They confirmed this hypothesis by showing that overexpression of hudA in a hudR-mutant background resulted in attenuated virulence of P. aeruginosa in flies and that hudA hudR double mutants had normal virulence. Flies have also been used to differentiate virulence of P. aeruginosa strains, such as those isolated from the sputum of cystic fibrosis patients (who are particularly sensitive to infection by this pathogen) (Lutter et al., 2008; Salunkhe et al., 2005; Sibley et al., 2008) . For these assays, flies are either fed different strains of bacteria obtained from burn wounds or from cystic fibrosis patients, or bacteria are inoculated by wounding flies. Similar infection experiments can be performed to identify interactions between P. aeruginosa and other microbes present in sputum that could contribute to the virulence of this pathogen (Sibley et al., 2008) . Virulence factors of human fungal pathogens can also be identified in Drosophila (Ben-Ami et al., 2010; Chamilos et al., 2010; Chamilos et al., 2008; Lamaris et al., 2007) . For example, gliotoxin produced by the filamentous fungus Aspergillus fumigatus is required for virulence of this pathogen in both flies and mice (Spikes et al., 2008) , and Cas5 has been shown to be a transcription factor regulating a set of genes required for integrity of the cell wall of C. albicans . Adult flies have also been used as an intact organism to screen for drugs that block fungal infection (Chamilos et al., 2006a; Chamilos et al., 2006b; Lamaris et al., 2009; Lamaris et al., 2008; Lionakis et al., 2005) . In the previous section, we discussed strategies by which flies can be used to screen for pathogen virulence factors. In this final section, we consider the advantages of Drosophila as a model for analyzing virulence factor function, and for identifying the host proteins and pathways that they target (e.g. Fig. 2C ). Although cellbased expression systems and biochemical experiments performed with purified virulence factors can be invaluable for establishing mechanism of action, they do not necessarily predict how such factors will act in an infected organism -either systemically or in selected tissues, in which cell-autonomous and non-cellautonomous processes might be important. Model systems such as flies and worms are ideal for this level of analysis owing to the great variety of genetic tools available to tease apart the effects that such factors might have on specific host pathways and biological processes. Although flies and worms are only distantly related to humans, many virulence factors target host proteins and pathways that are among the most conserved in eukaryotes -thus, studying the effect of pathogens in these organisms is often highly relevant to human disease. In addition, as discussed below, studies in model organisms also enable examination of the combinatorial effect of two or more virulence factors, which is more challenging in intact mammals. Finally, we highlight in Box 1 how studying the effect of toxins can shed light on basic cellular processes. Drosophila is an excellent in vivo genetic system for analyzing toxin activities in a multicellular and organ context given the highly conserved nature of many host targets of these virulence factors. For example, flies have been used to study the activity of the virulence factor ExoS from P. aeruginosa, which encodes a factor One of the first uses of toxins in flies was to genetically ablate specific cells with cell-lethal toxins such as diphtheria toxin (Kunes and Steller, 1991) or ricin (Moffat et al., 1992) . It is also possible to block the neuronal activity of cells without killing them, as with tetanus toxin (TTX), which was used to block synaptic transmission in the nervous system (Allen et al., 1999; Baines et al., 1999; Reddy et al., 1997; Sweeney et al., 1995) and activity-dependent regulation of synaptic size and function (Nakayama et al., 2006) . TTX has been used in a myriad of Drosophila studies to inhibit neurotransmission in various processes, including learning and memory, locomotion and courtship (for a review, see Martin et al., 2002) , circadian rhythms (Johard et al., 2009; Kaneko et al., 2000) , and the serotonin-dependent response to light (Rodriguez Moncalvo and Campos, 2009) . Similarly, application of cholera toxin (CTX), an ADP-ribosylation factor, was used to study the function of the G protein Concertina, which is involved in initiating embryonic gastrulation (Morize et al., 1998) . Similarly, transgenic expression of a UAS-CTX-A construct helped to distinguish which heteromeric G proteins contribute to wing maturation (Katanayeva et al., 2010) . Indeed, these and other toxins, which neutralize or alter the activities of multiple host proteins, can be used to perform a variety of in vivo pharmacological studies to complement classical genetic loss-offunction studies. containing a domain with Rho-GAP activity (which can inactivate host small GTPases of the Rho/Rac subfamily). During P. aeruginosa infection, ExoS is injected into host cells by a type-II secretion system (TTSS), and infection of flies with P. aeruginosa leads to rapid death that depends on TTSS function (Fauvarque et al., 2002) . When the GAP domain of ExoS (ExoSGAP) is expressed in fly hemocytes, phagocytosis is inhibited (Avet-Rochex et al., 2005) . In addition, expression of ExoSGAP in flies increases their sensitivity to infection by P. aeruginosa (Avet-Rochex et al., 2005) , and this effect can be rescued by co-expressing host Rac2 with ExoSGAP (Avet-Rochex et al., 2007) . These studies provide evidence that host Rac2 is inhibited by bacterial ExoSGAP during infection. A second example illustrating the utility of Drosophila for investigating toxin activities in vivo is provided by studies of Helicobacter pylori (Fig. 2C) , which is associated with the development of gastric ulcers and cancer in humans. Under normal circumstances, ligand-initiated receptor tyrosine kinase (RTK) signaling in both fly and mammalian cells is mediated by a receptorassociated protein complex including Grb2 (Drk), Gab (Dos) and Shp-2 [Corkscrew (Csw)] (Drosophila protein names are shown in parentheses) that then activates signaling via the downstream components of the Ras-MAPK pathway. Drosophila played a prominent role in discovering key components of this pathway and in establishing the order of molecular events that take place during signaling (Simon, 2000) . In mammalian cells, the H. pylori virulence factor CagA activates RTK signaling at the level of SHP-2, a tyrosine phosphatase that is homologous to Drosophila Csw (Hatakeyama, 2008; Hatakeyama, 2009) , which acts downstream of Gab (Dos in flies) (Herbst et al., 1996; Raabe et al., 1996) . Studies in Drosophila confirmed the hypothesis that CagA can bypass the need for signaldependent activation of Dos in an intact organism, because CagA expression in Drosophila embryos or in the adult eye was capable of rescuing dos-mutant phenotypes (Botham et al., 2008) . Furthermore, the ability to activate effectors of the Sevenless RTK pathway in the eye was shown to be dependent on the downstream effector Csw, validating the hypothesized role of CagA in the RTK signaling pathway acting between Gab (Dos) and SHP-2 (Csw). Beyond providing a multicellular model for assigning known biochemical activities to virulence factors, Drosophila can also be used as a tool to discover completely new activities of virulence factors. For example, Bacillus anthracis, the etiological agent of anthrax, produces two toxic factors required for systemic virulence (Lacy and Stevens, 1999; Mourez, 2004; Tournier et al., 2007; Guichard et al., 2011) : lethal factor (LF), a zinc metalloprotease that cleaves MAPKKs (Duesbery et al., 1998; Vitale et al., 1998) , and edema factor (EF), a highly active calmodulin-dependent adenylate cyclase (Leppla, 1982) . Both LF and EF are essential for the lethal effects of anthrax (Pezard et al., 1991) , which culminates in vascular failure and septic-shock-like death. An important unanswered question is, how do LF and EF, with such seemingly disparate enzymatic activities, collaborate during infection (particularly within vascular endothelial cells, which become leaky at advanced stages of disease, leading to death)? In initial studies, we showed that anthrax toxins act on Drosophila homologs of their known targets in mammalian cells (Guichard et al., 2006) . In addition to these known effects of LF and EF, we observed that both toxins also caused adult wing and bristle phenotypes similar to those caused by inhibition of the Notch signaling pathway, and blocked expression of Notch target genes in developing wing imaginal discs (Guichard et al., 2010) . Moreover, these toxins interacted in a synergistic fashion to block Notch signaling (Fig. 3A) . Further analysis of this Notch-like phenotype revealed that it resulted from failure to recycle the Notch ligand Delta to the cell surface (Guichard et al., 2010) . EF was found to reduce the levels and activity of the small GTPase Rab11, whereas LF reduced cell surface levels of the Rab11 binding partner, Sec15 (Fig. 3B ). Sec15 is part of an octameric protein complex known as the exocyst, which targets proteins, including Delta and the cell adhesion molecule DEcadherin, to adherens junctions (accordingly, DE-cadherin trafficking to adherens junctions was also reduced by EF and LF). These results from flies were validated in human vascular and lung endothelial cells by our collaborators in Victor Nizet's laboratory (Fig. 3C) , who also showed that EF reduced epithelial barrier integrity in a cell culture assay and in vivo in mice (Guichard et al., 2010) . Maintenance of vascular integrity depends on cell-cell adhesion (Dejana et al., 2009) , and cell-cell communication mediated by Notch signaling plays a role in promoting the formation of primary (or patent) vessels over more permeable microvessels (Hellstrom et al., 2007; Leslie et al., 2007; Lobov et al., 2007; Roca and Adams, 2007; Siekmann and Lawson, 2007; Suchting et al., 2007) . By inhibiting these two interrelated processes, and possibly interactions between endothelial cells and other vascular cell types such as mural cells, anthrax toxins might contribute to the latestage effects of anthrax infection when disruption of endothelial barrier function leads to lethal vascular collapse. Once sufficient levels of anthrax toxins are produced, they can be fatal even if the bacterial infection is eliminated with antibiotic treatment. Thus, these studies of anthrax toxins initiated in flies and validated in mammalian models might ultimately have therapeutic implication for treating humans infected with anthrax or for other conditions compromising vascular integrity. Given the compact sizes of viral genomes, only few viral proteins fall into the category of bona fide virulence factors, similar to the potent bacterial toxins discussed above. By contrast, most viral proteins are dedicated to basic processes essential to the virus life cycle, such as entry or exit, replication, or manipulation of host processes such as transcription or translation. Model organisms are useful for examining specific interactions between viral and host proteins to gain insights into their mechanisms of action. An excellent example of using the full complement of Drosophila tools to study a viral pathogen was carried out by Cherry and colleagues, who showed that fly cells can be infected with VSV. VSV can replicate in these cells to generate mature viral particles that can infect mammalian cells. They showed that infection of adult flies with VSV induces autophagy (Shelly et al., 2009) and that autophagy was mediated by VSV-G, a pathogen surface protein that is recognized by Drosophila cells. The authors found that induction of autophagy plays an important role in protecting against VSV infection and then asked what host pathways might mediate the autophagy response to VSV. A variety of elegant genetic epistasis experiments demonstrated that the PI3K-Akt pathway was attenuated by VSV infection, thereby relieving its constitutive (Guichard et al., 2010) . WT, wild type. (B)Analyze mechanisms of toxin action. The Notch-like phenotypes caused by expression of LF or EF in the wing both result from inhibition of endocytic recycling of membrane cargo to the AJ by the exocyst complex. EF acts by reducing the levels and activity of the Rab11 GTPase, which indirectly results in a loss of large vesicles containing its binding partner Sec15-GFP, a component of the exocyst complex. LF does not seem to alter Rab11 levels or function, but inhibits the formation of large Sec15 vesicles (Guichard et al., 2010) . (C)Validate toxin mechanism in vertebrates. Human brain microvascular endothelial cells were treated with purified EF toxin or LF toxin. As in fly cells, both toxins greatly reduce the number of Sec15-GFP vesicles in these cells and reduce cadherin expression (Guichard et al., 2010) . (D)Examine interactions between toxins. Cooperative interactions between toxins or other virulence factors can be assessed by co-expressing them in specific cells and comparing the effects of both toxins to that of the action of either toxin alone. In the example shown, anthrax toxins were expressed alone or in combination using a weak GAL4 driver to express low levels of the toxins. Each panel consists of an adult wing (top) and a larval wing imaginal disc showing expression of the Notch target gene wingless (wg) along the future edge of wing in third instar larvae (bottom). Expression of LF or EF alone (+LF or +EF, respectively) has little or no effect on formation of the wing margin (compared with WT). When LF and EF are co-expressed, the wing margin virtually disappears, as does expression of wg along the primordium of the wing margin. (E)In vivo structure-function analysis of toxins. The systemic activities of mutant forms of toxins or other virulence factors can be assessed in Drosophila. Such activities include cell-non-autonomous effects mediated by intercellular signaling systems, which are difficult to screen for in cell culture. In the simple case shown in this panel, high levels of LF expression lead to reduced wing size (middle panel) and a single point mutation in the LF catalytic domain renders it inactive (right panel). Panels A-D adapted from Guichard et al. (Guichard et al., 2010) with permission. Panel E adapted from Guichard et al. (Guichard et al., 2006) , with permission. Host-pathogen interactions in Drosophila PERSPECTIVE inhibition of autophagy ( Fig. 2A) . Akt activation is also attenuated by expression of the SARS-Coronovirus Membrane protein in flies, which in this case results in increased apoptosis (Chan et al., 2007) . In another study, host factors required for the HIV accessory protein Nef to downregulate expression of the human CD4 protein were identified by RNAi screening in Drosophila S2 cells expressing human CD4. These factors included components of the clathrinassociated AP2 complex, which was then validated as an essential cellular component mediating a similar Nef-CD4 interaction in human cells (Chaudhuri et al., 2007) . Classic genetic approaches in Drosophila can also be applied to probing structure-function relationships of toxins or other virulence factors. One straightforward approach to define domains of a toxin that are important for producing the phenotype of interest is to mutagenize flies carrying a UAS-toxin construct and to screen for loss of the phenotype that results from expression of the wild-type toxin (Fig. 3E ). One can then PCR amplify the mutated UAStransgenes and sequence the putative mutant allele to determine the molecular nature of the loss-of-function mutation. It is also possible to screen for mutations in the transgene that results in altered phenotypes caused by dominant gain-of-function mutations (Guichard et al., 2002) . An important goal of this review has been to convince readers from other fields that flies provide a broad range of advantages for studying host-pathogen interactions at the level of the cell, tissue, organ and intact organism. As discussed, genome-wide RNAi screens in Drosophila cell culture have generated a wealth of new information regarding the genes involved in mediating basic host cellular responses to pathogens, such as those involved in innate immunity, phagocytosis and restriction of intracellular pathogen survival. These cellular studies can be complemented by studies that aim to identify host resistance factors and pathogen virulence factors using intact flies as infection models. Studies in flies also provide the potential to explore mutualistic interactions with intracellular endosymbionts, and to conduct mechanistic analysis of specific virulence factors, and combinations of these factors, using the state-of-the-art genetic tools available in Drosophila. A particular advantage of model systems such as yeast, C. elegans and Drosophila is the potential to examine arrays of genetic combinations to identify factors produced by the host or pathogen that act redundantly (host) or that genetically interact (host or pathogen). These types of studies are inherently number intensive (known as 'the n problem'), because many combinations must be analyzed in a comprehensive fashion. However, there are excellent examples in which combinatorial genetic analysis has been used to investigate cellular processes involved in other types of human disease. For example, the interacting components of the DNA mismatch repair machinery were first identified in yeast, and the same components were found to interact in humans in a dominant manner and to contribute to cancer (Kolodner, 1995) . Drosophila cells and intact fly mis-expression systems (e.g. combined RNAi expression) are also well suited for such analyses, which would be prohibitively expensive and labor intensive in vertebrate models. It is of course important to validate results obtained in single cells or invertebrate model systems in vertebrates, a process previously referred to as 'closing-the-loop' (Bier, 2005) . It is possible to envision a tiered system of analysis in which initial discoveries that are made using powerful model genetic systems, including yeast, worms and flies (in cells and in whole organisms), are then validated in vertebrate models, including zebrafish, mice and human cells, and finally are linked via human genetics to specific disease processes. For example, in a recent study of genes on human chromosome 21 causing congenital heart defects when overexpressed in individuals with Down syndrome, a combined genetic analysis in flies, mice and humans pointed to two interacting genes, DSCAM and COL6A2, as contributing to formation of atrial septal defects (Grossman et al., 2011) . In addition to assessing combinatorial contributions of host factors, Drosophila is well suited for examining cooperative interactions between pathogen virulence factors, as presented in the examples above. Many pathogens produce a complex cocktail of virulence factors, subsets of which are often co-expressed from neighboring genes in so-called pathogenicity islands. These coregulated virulence factors are typically delivered by a dedicated injection system and often act by unknown means in various combinations in different host cell types. Such virulence factors from a given pathogenicity island can be expressed in various combinations in specific cell types to identify specific cellular contexts in which they interact. Given the great success of the fly for analyzing the activities of single host or pathogenic factors in disease processes, it will interesting to see whether it also serves as a robust system to study more complex networks of interactions between host pathways or pathogen virulence factors. With the advent of whole genome RNAi tools and comprehensive mutant collections, flies should also provide an important intact model system for identifying unknown activities of virulence factors that act in a multicellular context to inhibit specific signaling systems or to alter contact between neighboring cells in structured tissues and organs. Combined use of Drosophila cells and intact flies in moderate-to high-throughput drug screens is also emerging as an effective strategy to identify compounds, or combinations of existing compounds, that alter the activity of host pathways to counter the effect of pathogens. Clearly, flies have a bright future as tools for further deconstructing human host-pathogen interactions.

The Organisation of Ebola Virus Reveals a Capacity for Extensive, Modular Polyploidy

BACKGROUND: Filoviruses, including Ebola virus, are unusual in being filamentous animal viruses. Structural data on the arrangement, stoichiometry and organisation of the component molecules of filoviruses has until now been lacking, partially due to the need to work under level 4 biological containment. The present study provides unique insights into the structure of this deadly pathogen. METHODOLOGY AND PRINCIPAL FINDINGS: We have investigated the structure of Ebola virus using a combination of cryo-electron microscopy, cryo-electron tomography, sub-tomogram averaging, and single particle image processing. Here we report the three-dimensional structure and architecture of Ebola virus and establish that multiple copies of the RNA genome can be packaged to produce polyploid virus particles, through an extreme degree of length polymorphism. We show that the helical Ebola virus inner nucleocapsid containing RNA and nucleoprotein is stabilized by an outer layer of VP24-VP35 bridges. Elucidation of the structure of the membrane-associated glycoprotein in its native state indicates that the putative receptor-binding site is occluded within the molecule, while a major neutralizing epitope is exposed on its surface proximal to the viral envelope. The matrix protein VP40 forms a regular lattice within the envelope, although its contacts with the nucleocapsid are irregular. CONCLUSIONS: The results of this study demonstrate a modular organization in Ebola virus that accommodates a well-ordered, symmetrical nucleocapsid within a flexible, tubular membrane envelope.

Viruses have evolved as genome packaging machines to efficiently transfer nucleic acids between susceptible host cells, ensuring replication. The majority of viruses have hollow, quasispherical shells rather than tubular structures, perhaps because this gives the most efficient packaging of nucleic acid with a fixed copy number of coat protein subunits. In non-enveloped viruses, the volume enclosed by the (usually) icosahedral structure is a constraint on the size of the genome, giving a limited capacity to encode capsid proteins, and usually restricts the genome copy number, or ploidy of the virion, to one [1] . Most membraneenveloped viruses are also quasi-spherical, but their symmetry is frequently less well-ordered, which is usually described as pleomorphic. This feature allows some flexibility in volume, which could accommodate variation in the size of the genome or its copy number. Nevertheless, most viruses, irrespective of their architecture, appear to have evolved to encapsidate only a single copy of their genome within the protein or protein/lipid shell, or a dimeric copy in retroviruses. Notable exceptions are the Paramyxoviridae and the Birnaviridae where particles may contain up to four copies of the RNA genomes [2, 3] . Although some strains of influenza can produce elongated virions, there is a mechanism that selectively encapsidates only one set of genome segments in each virion [4] . The Filoviridae family, including the Ebolavirus and Marburgvirus genera, cause haemorrhagic fevers with high mortality in humans, and no effective treatments are currently approved [5] , although candidate vaccines are promising [6] . The 18.9 kb single-stranded negative-sense non-segmented RNA genome of Ebola virus (EBOV) codes for at least eight proteins. The ribonucleoprotein complex is composed of the nucleoprotein (NP), polymerase protein (L), VP24, VP30, and VP35. The trimeric transmembrane glycoprotein (GP) forms surface spikes on the virion envelope and also has a soluble form, while the matrix protein, VP40, is associated with the inner surface. [5, [7] [8] [9] . The GP spike, a class I fusion protein, mediates cellular attachment and entry and is extensively glycosylated, especially in the glycan-rich mucin-like domain [10] [11] [12] [13] . Three proteins, VP24, VP35 and NP are essential for nucleocapsid formation [14] . Although some of the major protein interactions that occur during EBOV morphogenesis have been characterised [14, 15] , the three-dimensional (3D) structure and molecular arrangements have not been previously determined. Structural details are essential to understand how protection of the genome, cell binding, entry, and immune evasion are achieved in a filamentous animal virus, and to determine how this unique morphology plays a role in pathogenesis. Research on filoviruses has been hampered by their status as biosafety level 4 pathogens. Previous investigations of filovirus structures within embedded, sectioned and metal-stained cells by electron tomography revealed few details of the high resolution oligomeric structure [16, 17] . It has been demonstrated that aldehyde-fixation alone, and subsequent cryo-electron microscopic imaging in the frozen-hydrated state preserves structures, at least up to 12 angstroms resolution [18] , and in some cases, fixation improves the resolution achievable [19] . In addition, it has also been shown that high-resolution X-ray structures can also be obtained in the presence of aldehyde fixatives [20] . Therefore, we analyzed purified and isolated EBOV and Ebola virus-like structures using cryo-electron microscopy (cryo-EM), and cryoelectron tomography (cryo-ET). In the current study, the Zaire strain of EBOV was purified and inactivated by paraformaldehyde fixation: excess fixative was then removed by dialysis to reduce beam damage for imaging in the frozen-hydrated state. The flashfreezing at liquid ethane temperatures used in cryo-electron microscopy preserves the structural and molecular detail, avoiding artifacts associated with conventional EM methods, such as dehydration and/or sectioning or staining, that usually prevent detailed structural analysis. Digital image processing reveals the 3D organization of EBOV, including the structural arrangement of component molecules at resolutions of 14-19 Å . We identified filamentous EBOV particles 20 microns or longer, with a well ordered internal structure, and a helical nucleocapsid giving an internal ''herring-bone'' appearance using cryo-EM and cryo-ET (Figures 1, 2 , S1, S5). The nucleocapsid, as observed within intact viral particles, has a uniform helical structure (Figures 1, 2, 3, 4) and is enveloped by a membrane coated by an external layer of GP spikes. From the same image data set, we combined extracted volumes from tomograms with 2-D single particle processing to determine the structure of the GP spikes ( Figure 5 ) to a resolution of 14 Å as measured by the Fourier Shell Correlation (FSC) 0.5 criterion. Virions are rarely straight. Variation in the overall length of virions is non-random, and they fit into ordered size classes ( Figure 1 ). Analysis of 2090 distinct intact virions with a nucleocapsid from cryo-electron micrographs shows that the most common class length (53%) of virus particles is 982679 nm ( Figure 1A , Table S1 ). The other size classes are multiples of this length. Enveloped filovirus particles have several Empty and linked EBOV structures were excluded from the histogram data. A single G1-single/comma shaped EBOV is shown (inset on the right, G1 = 1 copy of genome). (B) Low magnification cryo-images showing: G1-single/comma shape, G1-single/ linear, G5-continuous (G5 = 5 copies of genome). (C) High magnification of a G1-(single genome) virion with a region filtered to emphasize the nucleocapsid. (D) Low magnification image of a G4-linked EBOV, each genome copy is indicated and numbered, the red arrows show the transition points between nucleocapsids. The circular holes (filled with vitreous ice) appear as lighter regions and the support film (''quantifoil'') appears dark grey. A ''linker'' region is shown at higher magnification (inset). doi:10.1371/journal.pone.0029608.g001 different morphologies (shown in Figures 6, S1 , S5). These configurations are ''single'' particles, containing a nucleocapsid of uniform length, (which we postulate to contain one copy of the genome), ''continuous'' particles, with nucleocapsids of a length of the single virion multiplied by an integer of 2 or greater; and ''linked'' virions composed of a series single-genome nucleocapsids connected by short sections of empty envelope. Negative staining can cause drying and staining artefacts which hamper accurate measurements and preclude 3D analysis. Nevertheless, in previous studies using this technique, the average length of Marburg virus Figure 2 . Image processing of Ebola virus. Linear 2D averaging of EBOV: the envelope and nucleocapsid are prominent features (A). The line trace is colour-coded as follows: red, spike; beige, lipid envelope; green, membrane-associated proteins; white, membrane-nucleocapsid gap; blue and purple, outer and inner nucleocapsid. (B) 2D class averages of envelope plus inner face. (C) VP40 VLPs, showing 2D averages from the from side regions (first two) and end-on/central regions (last three). In (A-C) representative individual repeats have been highlighted in color using the same scheme as in (A). (D) Schematic model of the nucleocapsid and envelope, highlighting the relative distribution of NP to VP40. (E,F) 3D reconstruction of the nucleocapsid with the same colour scheme as in (A). The location of the inner nucleocapsid, and the bridge are indicated. The reconstruction is presented at a volume threshold that would encompass a single copy of each of these proteins, and the viral RNA. In (E) the vertical (protein-protein) and horizontal (protein-RNA) contacts are indicated by yellow and white arrows, respectively. (G) Various recombinant nucleocapsid-like structures, and authentic EBOV, which have been studied by electron microscopy [14, 15, 21, 28] . 3D schematics of these structures highlighting the RNA and protein composition and the diameter of these structures, at the same scale for comparison to (E). doi:10.1371/journal.pone.0029608.g002 was measured as 790 nm, and EBOV as 970 nm [21, 22] : the latter is very similar to the mean length of 982 nm measured in the current study by cryo-TEM of frozen hydrated specimens. In addition, discrete sub-populations of virions of double or triple the average length were also previously reported in centrifuged virus preparations [21] . We also observed ''empty'' filaments that lack nucleocapsids, with a random length and a smaller diameter ( Figure S1C ). These empty nucleocapsids are also visible in previously published thinsection micrographs of human EBOV infected pathology specimens, e.g. [23] , and are thus probably not an artifact of cell culture. It is also unlikely that the polyploidy we have observed in EBOV is a peculiarity of the particular viral isolate or of the Vero cell line that we used for this investigation. Previous studies using negative stain EM have shown that both Marburg virus and several different EBOV species can produce filamentous virions up to 14 mm long when grown in several different cell lines [21, 22] . Blood specimens from Guinea-pigs and monkeys that were inoculated with primary isolates of the Marburg agent, when centrifuged and observed by negative stain, also showed filamentous virions of variable length, with the average reported as about 1 mm in length but with a smaller proportion being over 2 mm in length [24] . In addition, viral filaments of lengths from 1.4 to 1.6 mm long, and occasionally as long as 2.6 mm, can also be seen in published ultrathin sections of EBOV infected human and monkey tissues [23, 25] . Thin-section EM always underestimates the lengths of virions, since filament profiles are frequently truncated by the section plane: in addition, tissue shrinkage of 10 to 20% during dehydration and resin embedding is common. Thus, it is probable that polyploid EBOV with multiple nucleocapsids are also produced in naturally infected humans and animals, though it is not possible to make a direct comparison with cell cultured virus: it would be very difficult to obtain large enough quantities of concentrated virus from animals to carry out detailed cryo-TEM analysis and measurements. The Ebola nucleocapsid structure was solved to 19 Å resolution (FSC 0.5 criteria) using linear regions of 34,605 images (taken at 3-4 mm defocus, Figure 2 ). Since virus particles were not straight enough for conventional helical image processing, a combination of tomography, sub-tomogram averaging, single particle averaging, and the iterative helical real-space construction method were used [26, 27] . The EBOV nucleocapsid is a right-handed doublelayered helix with an outer diameter of 41 nm and a hollow inner channel 16 nm in diameter as determined by image analysis and tomography (Figures 2, 3 , 4, S2, S4, movie S1). The pitch is 6.96 nm, with 10.81 repeats per helical turn ( Figure 2 ). The inner nucleocapsid, composed of large subunits, is linked by vertical and horizontal contacts between the large subunits ( Figure 2E ,F). The horizontal contacts occur between the large subunits at a diameter of 22.3 nm. The vertical contacts linking the coils have a higher density than the horizontal ones, so we interpret the horizontal contacts as involving viral RNA (white arrow, Figures 2E, 4E ; movie S1) and the vertical contacts as protein-protein interactions (yellow arrow, Figure 2E ). The NP subunits are also linked by an outer horizontal layer at a diameter of 37 nm. This layer consists of a ring of bridges between adjacent large subunits ( Figure 2 ). The bridges joining the NP subunits are composed of two lobes, one of which of which is slightly bigger than the other. Previous studies that produced recombinant nucleocapsid-like structures showed that expressed VP24 and VP35 both independently associate with NP, but that all three proteins together are necessary to produce ,50 nm diameter helical nucleocapsid-like structures. When VP35, VP30, VP24, and NP were transfected together, approximately 50 nm diameter helical nucleocapsid-like structure was also generated, whereas NP alone generated helical NP-RNA complexes ,20-25 nm in diameter, which were nuclease sensitive [14, 28, 29] . Taken together these results suggest that VP24 and VP35 are the structural components of the bridge located on the periphery of the nucleocapsid ( Figure 2G ). It is not possible to accurately delineate VP24 and VP35 within the bridge at this resolution, however the density of the bridge is consistent with a predicted total mass of 60 kD, thus each bridge is composed of one molecule of VP24 and one molecule of VP35. It is likely that the larger lobe is VP35 and that VP24 therefore resides within the smaller lobe ( Figure 2F ). Thus, each bridge is composed of a VP24-VP35 heterodimer that holds adjacent NP molecules together horizontally. This structure explains how VP24 and VP35 are able to independently interact with NP, and why all three are required for the formation of a double-layered nucleocapsid. In addition, it implies that VP24 and VP35 can interact with each other as well as each interacting with a different site on the NP molecule in order to make an oligomeric structure. The recombinant nucleocapsid data indicates that both VP35-VP30-VP24-NP and VP35-VP24-NP produce approximately 50 nm diameter helical structures that are indistinguishable from nucleocapsids produced by EBOV. This indicates that VP30 does not increase the diameter of the nucleocapsid. We propose that VP30 lies in the interior of the nucleocapsid and is not part of the bridge on the periphery of the nucleocapsid. This localization is consistent with previous work showing that VP30 is a component of the nucleocapsid, and associates with NP, but is non-essential for nucleocapsid formation [14, 30, 31] . Our model, in which the inner layer at 22.3 nm diameter is RNA-NP, and the outer bridge centered at 37 nm is composed of VP24-VP35 heterodimers, with VP30 bound to NP, is thus consistent with these previous observations [14, 28, 29] , and suggests that the outer VP24-VP35 heterodimer bridge functions in the stabilization and/or protection of the nucleocapsid. We have modeled the arrangement of the RNA within the EBOV nucleocapsid (Table S2 ). Since no atomic resolution data on the ribonucleoprotein structure of filoviruses have yet been reported, the ribonucleoprotein ring-structures of other members of the mononegovirales are useful for comparison. Although the mass of the nucleoproteins, and the number of nucleotides per subunit differs amongst these virus families, a model of the 18.9 kb EBOV genome with the RNA following a circular fixed radius as in respiratory syncytial virus (RSV) [32] would make a nucleocapsid containing one copy of the EBOV genome about 914 nm long (Table S2 ). This fits closely with the 982 nm length class (53% of the particles observed in this study) containing a single copy of the EBOV genome threaded through the NP at a fixed radius. Allowing 33 nm of space for the membrane envelope to curve around at each end of the virus particle gives an estimated length for a single-genome virus particle of 980 nm, which is very close to that observed (982679 nm). The majority of the virus particles fall into size categories that are a multiple of the putative single genome length (G1), giving size classes of 1.960.15 mm (G2: 18.7% of the particles having 2 genomes); 2.960.2 mm (G3: 12% of particles having 3 genomes) and so on ( Figure 1A and Table S1 ) . The length of the longest particle measured was consistent with having 22 genome copies. Our model predicts a nucleotide to NP ratio of 13.0 (Table S2) , which is within the range of 12 to 15 nucleotides per NP molecule as measured by biochemical studies of Marburg virus [33] . In partially full virions, the membrane envelope is constricted at the transition point where the nucleocapsid ends and the empty membrane tube begins (Figures 1, S1C ). Empty virions ( Figure S1C ) have a similar structure to VP40-GP virus-like particles ( Figure 3A ). Both full and empty virions have a continuous layer of spikes projecting from the surface (Figures S1, S2, S3, movie S2), giving an overall diameter of 120 nm for full particles. The majority of virions are linear (Figures 1, S1A, S5 ), others have a ''comma-shaped'' appearance, with a globular head containing portions of the nucleocapsid that are curled-up or bent at one end ( Figures 1A,B and S1B). It is clear that ''toroidal'' virions previously identified by negative staining [22, 34] are a variation of comma-shaped virions. Internal vesicles of 20-40 nm in size are also sometimes observed at the ends of virions ( Figure S1D ). These vesicles appear to be formed during the process of envelopment of the nucleocapsid, since they were not observed in preparations of VP40 or VP40-GP VLPs. VP40 VLPs had wavy envelopes with an irregular diameter ranging between 48 nm and 142 nm (N = 49) compared to VP40-GP VLPs which were more ordered with a diameter between 50 nm and 91 nm (N = 26), (Figure 3 ). Thus the presence of GP, and certain contacts between the GP and VP40, play a part in stabilizing the tubular membrane envelope structure, and our observed structure agrees with previous reports that GP enhances VP40 VLP budding [35, 36] . The previously reported ''branched'', filamentous forms [34] were rarely observed: these consist of empty tubes (data not shown). It is possible that centrifugal virus purification disrupted most branched structures that were seen previously with negative staining of cell culture supernatant. The VP40 matrix protein shows a regular 5 nm lattice spacing ( Figure 2B , C). Both the nucleocapsid and the VP40 layers are ordered, however the contacts between them appear to be nonsymmetrical ( Figure 2D ), implying some flexibility in their intermolecular contacts. It has been shown that VP35 interacts with the VP40 and can be packaged into VP40 VLPs [37] . The localization of the VP24-VP35 bridges on the periphery of the nucleocapsid, may allow interactions of one or more of the nucleocapsid proteins with VP40, possibly through projecting low-density protein loops. There is a 6-7 nm gap of low density between the nucleocapsid and the VP40 layer, but tomography also shows discrete areas of connectivity between the nucleocapsid and matrix protein layers, which may be connections between the envelope and the nucleocapsid ( Figure 3I ). These results are substantiated by the analysis of sub-tomograms where helical symmetry is clearly evident but was not imposed (Figures 4, S4) . The use of sub-tomogram analysis improves the resolution of the data by averaging sub-tomograms together which are at different angular orientations. The helical nucleocapsid is clearly identified in the structure, including the gap between the envelope-VP40, and the nucleocapsid-VP40. The right-handed pitch of the helix is clearly discernable (Figure 4E ), as well as the putative location of several VP40 proteins, although the resolution of the tomographic data by itself is slightly lower than with single particle analysis. While showing the overall organization of EBOV, tomography also allows estimation of the stoichiometry of the major structural proteins (Figure 3 , movie S2). Both EBOV and VP40-GP VLPs have an irregular distribution of the GP spikes on the surface (Figures 3, S3) demonstrating the lack of any ordered lattice-like arrangement with the matrix protein VP40. The spikes are clustered, and the average centre-to-centre spacing is 15.2 nm ( Figure S3, and movie S3) . We calculate that a virion of 982 nm in length would have about 1888 copies of the GP spike, and 8391 copies of VP40. The wide spacing of the EBOV GP in the viral envelope allows plenty of space for free access and binding of any neutralizing antibodies directed at both the club-shaped head and stem region without stearic hindrance (Figures 3J, S3) . The nucleocapsid structure implies equimolar ratios of NP, VP24, VP30, and VP35. A genome of 18.9 kbp with 13 bases per NP with a single genome copy gives 1454 molecules of NP, VP24, VP30, and VP35 per virion. Previous analyses of Coomassie blue stained gels of purified EBOV predicted 625, 1208, 833, and 2686 protein molecules of NP, VP24, VP30, and VP35, per virion respectively [38] , which is in the same stoichiometric range as predicted by our model for the nucleocapsid, taking into account the variability of individual protein band staining by Coomassie blue, which is affected by factors such as distance of migration [39] , basic amino acid content [40] , and the extent of glycosylation [41] . Since NP is glycosylated, we anticipate underestimation of the NP content of virions [14, 41] . The GP spike is necessary for cellular attachment and fusion of EBOV. A definitive cell surface ligand for the receptor binding domain has not been identified, and a number of different cell surface proteins are able to enhance infection [42, 43] . Cleavage of GP results in two domains: GP1 containing the receptor binding domain and GP2 that contains the fusion and transmembrane domains. The structure of a smaller engineered fragment of GP1-GP2 has recently been determined by x-ray crystallography (3CSY pdb [42] ). We determined the structure of the entire EBOV GP trimeric spike at a resolution of 14 Å (FSC 0.5 criteria), by combining sub-tomograms from the spikes of VP40-GP VLPs (234 images) with EBOV images for the side perspective data (8084 images) using projection matching as previously described [44] ( Figure 5 and movie S4). In our reconstruction, the spike is in situ in the viral membrane, thus the transmembrane region and base of the spike, adjacent to the membrane, are less well defined than the distal region of the spike, due to the smaller differences in contrast between lipid and protein versus water and protein, as well as Fresnel fringes at the edge of the viral membrane. The spike extends 10 nm from the surface of the envelope, with a clubshaped head 6.5 nm in height, and a 3.5 nm long stalk. Docking of the previously determined X-ray structure into our cryo-EM map shows a good fit, with a correlation coefficient of 0.75 using the docking and correlation program SITUS [45] . The difference map calculated between our cryo-EM map and the GP1-GP2 structure identified volumes corresponding to the domains deleted to generate the 3CSY GP1-GP2 structure ( Figure 5 ). We show that the mucin-like domains (connected at V310 and E502 -shown in Green in Figure 5 ) completely fill the previously described bowllike chalice which contains the putative receptor binding sites [42] . The proximity of glycosylation sites in the GP1-GP2 structure suggests that the distal density of the spike contains the glycans that were deleted in order to construct the GP1-GP2 structure. In addition, each mucin-like domain has an ''arm-like'' projection, which extends radially at the distal end of the spike, to a maximum diameter of 13 nm. The localization of the mucin-like domain is consistent with previous studies showing that endosomal proteolysis plays a role in enhancing infectivity as well as binding of the Ebola GP to the plasma membrane [46, 47] . The other two major deletions in the 3CSY structure (N278-R299 and A189-Y214) are situated at the midpoint of the structure just above the stalk (shown in pink in Figure 5 ). Inclusion of the KZ52 Fab in the docked structure demonstrates that this neutralizing epitope (from a human survivor) is localized on the side of the stem region of EBOV GP trimer at the base of the clubshaped head, and that the Fab domain lies close to the lipid envelope when bound, and approximately tangential to the viral envelope. The densities putatively corresponding to N278-R299 and A189-Y214 are close to the KZ52 neutralizing site, but do not obstruct antibody binding. The mucin-like domains are out of the way and cannot interfere stearically with KZ52 Fab binding. This is consistent with previous results indicating that KZ52 binding does not require cathepsin cleavage [48] . We have thus delineated the low resolution structure of the glycocalyx or ''glycan cap'' that covers the distal end of the uncleaved EBOV GP spike, which is consistent with a proposed role in immune evasion [42] . Our data will enable docking of future structures to investigate receptor binding, antigenicity, and fusion mechanisms. We have shown that EBOV particles are capable of a high degree of polyploidy, made possible by the extreme length polymorphism of budding virus particles. Polyploidy in filoviruses may be more extensive than in any other virus family, with 46% of virions having more than one genome copy, and some having up to 22 copies. Polyploidy has been shown to increase infectivity rates in paramyxovirus and birnavirus [2, 3] . Attempts to investigate infectivity rates of EBOV by centrifugal fractionation of the different sized particles are stymied by the extreme filamentous morphology, as well as that fact that EBOV of different lengths have the same buoyant density (personal observations). A complex double layered helical nucleocapsid appears to be unique to filoviruses. In the case of rhabdoviruses, the bullet-shaped nucleocapsid precludes them from being linked sequentially. Although some influenza strains can produce filamentous virions, this morphology appears to be driven by the matrix protein only [49] . A filamentous morphology may have evolutionary implications by allowing genome length flexibility. It could also enhance the ability for viral dissemination in infected tissues, for example by diapedesis of budding filamentous virions through epithelial layers. Zaire Ebolavirus was propagated in Vero E6 cells and purified as previously described [50] . Ebola enriched samples were checked by SDS-Page and Western blotting, and rendered non-infectious by fixation with 4% paraformaldehyde. Excess fixative was removed by placing the fixed samples in a Slide-A-Lyzer G2 cassette with a 0.5 ml capacity, and a 10,000 MWCO (Thermo Scientific Pierce Protein Research Products, Rockford, Illinois, USA), followed by dialysis against PBS. Virus-like particles were produced as previously described [51] . All work with infectious Ebola virus (virus culture and purification) was performed in the biosafety level 4 laboratories at the National Microbiology Laboratory of the Public Health Agency of Canada, Winnipeg, Manitoba. Samples for cryo-electron microscopy (cryo-EM), and cryoelectron tomography (cryo-ET) were mixed with BSA coated Accessories, Wageningen, The Netherlands) at a ratio of 2:1 (virus:gold) for cryo-ET, and (9:1) for cryo-EM. Specimens (4 ml) were then applied to glow-discharged quantifoil grids with 2 mm holes spaced at 1 mm intervals (Quantifoil MicroTools GmbH, Jena, Germany). Grids were subsequently plunge cooled in liquid ethane using a Vitrobot Mark IV (FEI Company, Hillsboro, Oregon, USA). Specimens were transferred to a Tecnai 20 G2 transmission electron microscope (FEI) operated at 200 kV, equipped with a Gatan CT3500TR single tilt rotation lowtemperature specimen holder. For cryo-EM imaging was conducted at temperatures of ,2185uC. Images were recorded using an Eagle 4K CCD camera (FEI Company, Hillsboro, Oregon, USA). For single particle image analysis, images were taken at 50,0006 or 80,0006 magnification at 2-4 mm defocus, with a dose of 10 electrons/Å 2 . This corresponded to a pixel size at the CCD detector of 2.147 Å /pixel and 1.353 Å /pixel, respectively. For virus length measurements low magnification cryo-EM images were taken at 5,0006, 3,5006 and 2,5006. For cryo-ET single axis tomograms were taken at 25,0006, 29,0006 or 50,0006 magnification, at 28 m or 26 m defocus, with angle steps of 2u24u. Data were collected within tilt ranges of 660u, or 652u, with a total dose/tomographic data set of 47-60 electrons/Å 2 . For cryo-EM, data collection was done using the low-dose unit and software coupled with the TEM Imaging & Analysis (TIA) software (FEI Company, Hillsboro, Oregon, USA). Automated eucentricity determination, and focusing were performed using the Xplore3D data acquisition software (FEI Company, Hillsboro, Oregon, USA). For cryo-ET, data collection was done using the Xplore3D data acquisition software, the low-dose unit, and the TIA software (FEI Company, Hillsboro, Oregon, USA). The exact magnification in the microscope at the CCD detector was determined using a calibration grid (Pelco International, Redding, CA). Ebola virus length measurements (n = 2090) were made using the Image J software package [52] using the free hand line tool, and the analyse/measure function. For this analysis only viruses containing a continuous nucleocapsid were measured. Viruses with linked nucleocapsids and empty viruses were omitted. The measurements that were made in image J were then collated, analysed, and plotted, using Microsoft Excel. Tomographic image analysis of cryo-ET data was carried out with the Inspect3D Xpress software package (FEI Company, Hillsboro, Oregon, USA). The tomographic images were aligned to each other by a two-step process. The first step involved alignment of adjacent images by cross correlation. This process was repeated several times until the shift between adjacent images was below one pixel in either the X or Y plane. The second step involved the alignment of the entire image stack using 10 nm colloidal gold particles as fiducial markers. In this instance the term ''image stack'' refers to all the images collected in a single tomographic tilt. This procedure involves the selection of ten to twenty of the 10 nm gold particles and the subsequent identification and tracking of these particles in all of the images in the image stack. The locations of these particles are then used in conjecture with the tilt angles of each image to globally align all of the images to each other. The last step in this process was to calculate the three dimensional reconstruction of the tomogram from the aligned images. In this study we used the simultaneous iterative reconstruction technique (SIRT) algorithm with 10 iterations to calculate the final three-dimensional reconstruction (tomogram). Sub-tomogram image analysis of cryo-ET data was carried out with the Automated Recognition of Geometries, Objects, and Segmentations (ARGOS) software package (FEI Company, Hillsboro, Oregon, USA). For this analysis an 80 3 pixel subtomogram was extracted from a tomogram using the Chimera [53] software package. In this analysis the tomogram used contained a linear region of the Ebola virus ( Figure S4 ), and the 80 3 pixel sub-tomogram contained a single linear segment. This template was then used by the ARGOS software to conduct an exhaustive search of the original tomogram for similar structures. This analysis involved a six dimensional search matrix (three positional variants, and three rotational variants). The entire search process was sped up by the ARGOS software by utilizing parallel processing on the computer's graphics processing unit (GPU). Once individual sub-tomograms were selected based on correlation, they were inspected and compared to the initial template sub-tomogram. The extracted and aligned sub-tomograms were subsequently averaged with a filter that minimized the missing wedge artifact. This average structure then was used as the reference and the entire procedure was repeated several times. Single particle image analysis: software and hardware Single particle cryo-EM image processing was carried out using the EMAN/EMAN2 and SPIDER/WEB image processing program packages [54, 55] . Particle selection (EMAN) and contrast transfer function correction (EMAN2) were conducted on an Apple Inc. Mac Pro computer (12-core, Intel Xeon Nehalem processors 2.93 GHz, 32 GB Ram, Mac OS X 10.6.7). All subsequent calculations were performed on a Dell PowerEdge R900 4-way 64-bit Xeon X7460 processors, Six Core 2.67 GHz CPUs with 256 GB Ram running Linux (CentOS 5.2). Images were corrected for contrast transfer function (ctf) using the ''e2ctf.py'' function in the EMAN2 software package 5 , which estimates defocus and corrects for ctf by phase-flipping. Images of the spike (n = 8084 side perspective; n = 234 end-on perspective) and nucleocapsid (n = 34,605) were selected for image analysis. The resolution of the cryo-EM reconstruction was estimated by Fourier shell correlation using the FSC 0.5 criteria. In all subsequent sections image analysis procedures were conducted using the SPIDER software package unless otherwise stated. Analysis of initial images of the ''straight'' linear segments of the Ebola virus using Fourier transformation indicated that there was sufficient bending of the helical nucleocapsid to make standard helical analysis problematic. Therefore, an initial reference free single particle 2D analysis was conducted in EMAN using the ''startnrclasses'' program to identify any potentially recurring motifs within linear regions of the Ebola virus. In order to further investigate the nucleocapsid repeat identified in the 2D analysis the iterative helical real space reconstruction method (IHRSR) was implemented [26, 27] . This procedure requires an initial 3D helical reference structure which is used for image alignment. In this investigation a linear region of the Ebola virus which was extracted from a tomogram was used to generate this helical reference structure. This initial 3D structure was first pre-treated with a Gaussian mask to select only the nucleocapsid-containing region of the virus tomogram. An auto correlation function was then performed in which the volume was rotated around the helical axis, and translated along the axis of the nucleocapsid. At each rotational and translational position and autocorrelation value was calculated (between the shifted and unshifted volume). The net result of this process was the determination of the helical symmetry present in the tomogram. In order to analyse the data generated by this procedure the Microsoft excel spread sheet program was used. The correlation plots generated by this process solved the handedness, pitch, and number of repeats per turn for the nucleocapsid. These symmetry parameters were then applied to the tomogram to generate the initial 3D model. The IHRSR method was then applied to the 34,605 single particle images of the nucleocapsid as previously described [26, 27] . For the spike dataset two image populations were combined composed of side, and end-on perspectives. For the side view perspective images were subfield directly off of the Ebola virus cryo-EM images. For the end-on perspectives sub-tomographic volumes were extracted from the tomographic reconstructions of the Ebola VLP. The 3D volumes were then added in the ''Z'' plane to generate 2D projection averages which were then used for the subsequent single particle image analysis. The data were the processed using EMAN to generate an initial 3D reconstruction, which was then refined in SPIDER using the projection matching technique as previously described [44, 56] . The docking of the 3CSY.pdb [42] structure to the cryo-EM structure of the spike was accomplished using the SITUS [57] software package with the exception that only the GP1 and GP2 components of the 3CSY structure were used for the docking process. The ''floodfill'' program in SITUS was used to segment the spike component of the 3D cryo-EM reconstruction from the envelope component of the reconstruction. The segmented volume was then used for the docking procedure using the 'colores' function in SITUS. Once docked the entire 3CSY.pdb structure which included the Fab of the KZ52 neutralizing antibody was superposed over the docked GP1/GP2 component of the structure. The 3D cryo-EM reconstructions, cryo-ET reconstructions, 3D models of the Ebola virus, and the atomic resolution structure 3CSY.pdb were visualized using UCSF Chimera software package (Computer Graphics Laboratory, University of California, San Francisco, supported by NIH P41 RR-01081) [53] . The 3D images and movies presented in this manuscript were generated directly by UCSF Chimera software package. , Region three is shown, with the locations of correlation maxima shown with an X. The angular distance between each maximum was calculated from several plots. A total of 71 measurements gave an average angular distance of 33.6u+/ 28.5u between helical repeats, resulting in 10.7 repeats per turn. Using the 6.96 nm pitch (Fig. S8 ) the step in Z per helical repeat was calculated as 0.65 nm. These helical symmetry values were then imposed on the nucleocapsid tomogram and this structure was used as the initial reference volume for refinement using the iterative helical real space reconstruction method. (TIF) Figure S3 Surface spike distribution in the Ebola VLP. Longitudinal Z-slices through the top and middle of the particle are shown, as well as the end-on view (A). The tomogram is shown as a shaded surface at a density threshold that indicates the spikes (B). The volume from one side of the tomogram has been extracted, and a red-blue color scheme shows the depth at which the spikes are located. This region of the envelope has a surface area of 15,651 nm 2 . Selected spikes have been identified by red circles, the single particle reconstruction of the spike is shown at the same scale to the right in a red square for comparison. The same region in (B) is shown in (C) with a solid orange cylinder to provide a visual cue for the viral envelope. Eighty-six individual spikes were counted (white spheres) and have a patchy distribution (D), each spike would occupy an average area of 182 nm 2 , giving an average spacing between spikes of 15.2 nm. The reconstruction of the spike (blue) with the docked KZ52 Fab (purple) has been included to show that there is ample room for antibody attachment. (TIF) Figure S4 Extraction of Ebola nucleocapsid structure for sub-tomographic analysis. The tomogram of a linear region of the Ebola virus was used as the first reference for subtomogram analysis (A). When viewed along the helical axis (Y) or from the end perspectives (X,Z) the basic components are visible. The tomographic volume was also cylindrically masked along the X-axis, selecting only the density containing the nucleocapsid, to highlight the components of the nucleocapsid in the tomogram (B). Two-dimensional single particle image analysis was carried out with cryo-images (C) (not tomographic data sets), for comparison to the 3D tomographic data. The average shown in this panel was generated by reference free classification, using the ''startnrclasses'' program in EMAN [54] . The 6.96 nm helical pitch can be easily seen in the 2D average, but is also visible in the projections of the tomographic volume in (A, B) . (TIF) Figure S5 Representative low-magnification images of Ebola virus. Frozen hydrated virus is clearly visible with sections of the filamentous virus over both the support film and across the holes in the quantifoil film. Individual G1 (single genome copy) virus is circled in red, several sections containing a nucleocapsid are indicated by a blue arrowhead, and regions without a nucleocapsid are indicated by a magenta arrowhead. Globular heads are identified by yellow arrowheads. In this image the circles (light grey, 2 m diameter) are filled with frozen hydrated virus in a thin aqueous layer, and the quantifoil support film appears as darker grey. Table S1 Length analysis of ''continuous'' Ebola virus particles. The length of 2090 EBOV particles were measured using ImageJ [52] . The values in the ''model length'' column are based on multiples of the G1 mean length. The values in the in the ''mean length'' column were calculated directly from the data. Only full particles containing a continuously packaged nucleocapsid were measured, all others (linked-nucleocapsid and empty particles) were omitted from this analysis. The terms G1-G22 indicate the number of genomes/viral particle (i.e. G22 = 22 genomes). All measurements are in mm. Modeling of RNA in the Ebola nucleocapsid. Two previously determined atomic resolution structures of negative stranded RNA viruses (VSV (2GIC.pdb) [58] , and RSV (2WJ8.pdb) [32] ) were used to estimate the EBOV nucleocapsid length and number of nucleotides per nucleoprotein. Images of VSV (A) and RSV (B) are shown as a molecular surface with the protein in orange and the RNA as a green ribbon. From left to right, they show a surface view from the side, a side-on cross section, an end-on view, the RNA density alone in projection, and a rotational average of the projection. The VSV-based estimate, with a saw-tooth pattern of RNA in the helix, gave a nucleocapsid which 614.37 nm long, too short for the measured length of the G1 EBOV (982 nm). The RSV-based model, with a relatively straight/circular pattern of RNA in the helix predicted a nucleocapsid 914.55 nm long, which closely fits the measured length of G1 virions, after allowing ,34 nm space at each end to accommodate the curve of the envelope containing GP spikes and matrix proteins. The RSV-like model gives 13 nucleotides per nucleocapsid protein which is similar to previous biochemical estimates of 12-15 for Marburg virus [33] , suggesting that the RNA in the EBOV nucleocapsid is arranged in a smooth helical pattern at a diameter of ,22 nm. (TIF) Movie S1 3-D reconstruction of Ebola virus nucleocapsid. This movie shows the three-dimensional structure of the of the Ebola virus nucleocapsid as shaded surface representation. The surface is set at a density threshold which would include one copy of NP, VP24,VP30, VP35, and the RNA. The nucleocapsid rotates, and then is sliced through the Z-axis to show the internal components of the structure. Movie S3 Ebola virus spike distribution. This movie shows one surface of a cryo-electron tomogram of an Ebola viruslike particle, generated by expressing the VP40 and GP proteins. The structure rotates showing the distribution of spikes. The locations of individual spikes are identified by white spheres, and the reconstruction is replaced by a cylinder to show the patchy distribution of spikes on the surface of the virus-like particle. Movie S4 3-D reconstruction of Ebola virus spike. This movie shows the three-dimensional structure of the of the Ebola virus GP spike. The structure rotates showing views from different angles, and indicates the location of the docked 3CSY.pdb [42] structure with the GP1 and GP2 domains and KZ52 antibody (purple). (MOV)

3D QSAR Pharmacophore Modeling, in Silico Screening, and Density Functional Theory (DFT) Approaches for Identification of Human Chymase Inhibitors

Human chymase is a very important target for the treatment of cardiovascular diseases. Using a series of theoretical methods like pharmacophore modeling, database screening, molecular docking and Density Functional Theory (DFT) calculations, an investigation for identification of novel chymase inhibitors, and to specify the key factors crucial for the binding and interaction between chymase and inhibitors is performed. A highly correlating (r = 0.942) pharmacophore model (Hypo1) with two hydrogen bond acceptors, and three hydrophobic aromatic features is generated. After successfully validating “Hypo1”, it is further applied in database screening. Hit compounds are subjected to various drug-like filtrations and molecular docking studies. Finally, three structurally diverse compounds with high GOLD fitness scores and interactions with key active site amino acids are identified as potent chymase hits. Moreover, DFT study is performed which confirms very clear trends between electronic properties and inhibitory activity (IC(50)) data thus successfully validating “Hypo1” by DFT method. Therefore, this research exertion can be helpful in the development of new potent hits for chymase. In addition, the combinational use of docking, orbital energies and molecular electrostatic potential analysis is also demonstrated as a good endeavor to gain an insight into the interaction between chymase and inhibitors.

Raised blood pressure, especially systolic pressure (hypertension), is one of the striking factors inducing various diseases like heart failure, stroke, myocardial infarction and arterial aneurysm, and is a leading cause of chronic kidney failure [1] . A treatment of hypertension is to decrease the circulating volume and/or to slack the blood vessels [2] . Angiotensin II has important roles not only in the regulation of blood pressure but also in the development of vascular wall remodeling [3] . Conversion of angiotensin I (Ang I) to angiotensin II (Ang II) is catalyzed by well-known angiotensin-converting enzyme (ACE), which is a metallo-proteinase with dipeptidyl-carboxypeptidase activity. However, chymase (EC 3.4.21.39) which is a chymotrypsin-like enzyme expressed in the secretory granule of mast cells, also catalyzes the production of angiotensin II in vascular tissues even when ACE is blocked ( Figure 1 ). Chymase-dependent conversion of angiotensin I to angiotensin II and precursors of TGF-β and MMP-9 to their active forms. Chymase converts Ang I to Ang II with greater efficiency and selectivity than ACE [4] . The rate of this conversion by chymase is approximately four fold higher than ACE. Chymase shows enzymatic activity immediately after its release into the interstitial tissues at pH 7.4 following various stimuli in tissues. Since chymase has no enzymatic activity in normal tissues, chymase inhibitors are expected to have high safety because chymase inhibitors may not have an effect on any other targets in normal tissues [5] . In order to generate Ang II, human, monkey, dog and hamster chymases cleave the angiotensin I at Phe8-His9 peptide bond. Chymase also converts precursors of transforming growth factor-β (TGF-β) and matrix metalloproteinase (MMP)-9 to their active forms thus contributing to vascular response to injury. Both TGF-β and MMP-9 are involved in tissue inflammation and fibrosis, resulting in organ damage [6] . Previous studies have demonstrated the involvement of chymase in the escalation of dermatitis and chronic inflammation pursuing cardiac and pulmonary fibrosis [7] . Therefore, inhibition of chymase is likely to divulge therapeutic ways for the treatment of cardiovascular diseases, allergic inflammation, and fibrotic disorders. Chymase inhibition may also be useful for preventing the progression of type 2 diabetes, along with the prevention of diabetic retinopathy [8] . Moreover, the role of chymase in inflammation has prompted its restorative value in diseases such as chronic obstructive pulmonary disease (COPD) and asthma [9] . Chymase inhibitors are imperative for elucidation of the physiological functions of chymase and potentially useful therapeutic agents. Several chymase inhibitors such as sulfonyl fluoride derivatives [10] , Boc-Val-Pro-Phe-CO 2 Me [11] , Z-Ile-Glu-Pro-Phe-CO 2 Me, (F)-Phe-COGlu-Asp-ArgOMe [12] , N-(2-Naphthyl) carboxamido derivatives [13] , N- (2,2-dimethyl-3-(N-(4-cyanobenzoyl) amino) nonanoyl)-L-phenylalanine ethyl ester [14] , 3-benzylazetidine-2-one derivatives [15] , 1,3-diazetidine-2,4-dione derivatives [16] , methyllinderone derivatives [17] , chloromethyl ketone derivatives [18] , 1-oxacephem derivatives [19] , and 3-(phenylsulfonyl)-1-phenylimidazolidine-2,4-dione derivatives [20] have been reported previously. In general, chymase inhibitors readily decompose in plasma, thus the stability of the chymase inhibitors in human plasma has always been a matter of great concern. For a drug candidate, it is essential to enhance the stability of the active compound in human plasma. So, there is always a dire need to search for more stable inhibitors with high activity against human chymase. Many studies have indicated that computational approaches, such as predicting drug-target interaction networks [21] , prediction of body fluids [22] , predicting HIV cleavage sites in proteins [23, 24] , predicting protein metabolic stability [25] , predicting signal peptides [26] , identification of DNA Binding Proteins [27] , predicting the network of substrate-enzyme-product triads [28] , predicting protein subcellular locations [29, 30] , predicting proteases and their types [31] , predicting antimicrobial peptides [32] , predicting membrane proteins and their types [33] , predicting GPCRs and their types [34] , identifying nuclear receptor subfamilies [35] , predicting gram-negative bacterial protein cellular locations [36] , and predicting transcriptional activity of multiple site p53 mutants [37] , can provide many useful insights and data for which it would be time-consuming and costly to obtain by experiments alone. Actually, these data, combined with the information derived from the structural bioinformatics tools (see, e.g., [38] ), can timely provide very useful insights for both basic research and drug development. In view of this, the present study attempts to develop a new computational modeling method in the hopes it may become a useful tool for the drug development. A quantitative structure-activity relationship (QSAR) study is a helpful approach to quantitatively understand the relationships between molecular structures of inhibitors and their biological activities [39] [40] [41] [42] [43] [44] [45] [46] . Pharmacophore modeling and 3D-QSAR studies have been successfully applied previously for various drug discovery research, including glycoprotein (GP) IIb/IIIa antagonists, H 3 -antihistaminics, and dihydrofolate reductase inhibitors [47] [48] [49] [50] [51] . Electronic molecular features such as electron density, frontier molecular orbital density fields such as lowest unoccupied molecular orbital (LUMO), highest occupied molecular orbital (HOMO) and molecular electrostatic map have also been revealed to be significant in other QSAR studies to explain biological activity and molecular properties [52] . The HOMO density field was useful in a study of ACE inhibitors, and the LUMO density field was found to be important for explaining the TA100 mutagenicity [53, 54] . Thus, determining molecular electronic properties responsible for the potent activity of selected chymase inhibitors should illuminate the fundamental molecular level forces responsible for their potency. Various QSAR studies for chymase inhibitors have also been performed. The QSAR analysis of anhydride-type chymase inhibitors showed that aromatic substituents played an important role in determining the inhibitory potency of the compounds [55] . While, Hayashi and coworkers showed that introduction of various substituents in chloromethyl ketone derivatives resulted in a variation in their activity against human chymase [18] . A 3D QSAR model for the identification of stable chymase inhibitors has also been developed by Yuuki et al. 2003 [56] . The subject of the present study is to develop QSAR models and explore the key molecular features of chymase inhibitors influencing the protein-ligand binding and interaction, by exploring the dependence of inhibitory activities upon various physiochemical properties of these compounds. In order to accomplish these tasks, an exclusive computational strategy is applied by using various QSAR model building techniques such as pharmacophore modeling, molecular docking, and Density Functional Theory (DFT) (Figure 2 ). In the first phase of calculations, a pharmacophore model (Hypo1) comprising key chemical features for the identification of novel and diverse chymase inhibitors has been generated. After validation, this pharmacophore model is used as a 3D structural search query to find new classes of compounds with similar chemical features from chemical databases. The obtained hits are scrutinized based on their estimated activity and calculated drug-like properties. Molecular docking is also performed for the evaluation of compounds for important binding site interactions and affinity. Finally, we have carried out DFT-based QSAR studies on a set of chymase inhibitors retaining structural diversity and a wide biological activity range, along with potent hits retrieved by newly developed pharmacophore model (Hypo1). The objective of this DFT study is two-fold. One purpose is to derive the QSAR model itself and the other is to scrutinize the usefulness of conceptual DFT quantities. Moreover, it also served as a validation technique for the generated pharmacophore model. Various electronic properties such as LUMO, HOMO, and locations of molecular electrostatic potentials, are computed. The results of this study are expected to explore the crucial molecular features contributing to binding specificity and be useful for understanding the molecular mechanism by which these compounds act and can be further utilized to get compounds with better activity by rational modification. One of the main objectives of the present study is to generate a pharmacophore model for the identification of novel chymase inhibitors. To accomplish this, ten hypotheses with imperative statistical parameters were generated by HypoGen module of DS using a training set of 20 compounds (Figure 3 ). The hypotheses are generated with cost functions and correlation values by which they are estimated. The fixed cost, total cost and null cost values are calculated by HypoGen module during the hypotheses generation. The fixed cost is the lowest possible cost representing a hypothetically simplest model that fits all data perfectly, whereas the null cost value is equal to the maximum occurring error cost. For a more statistically significant hypothesis, there should be greater difference between these two cost values. The possibility of correlating the experimental and estimated activity data enhances to 75-90% with a cost difference of 40-60 bits between the total and null cost values [57, 58] . In the present work, the null cost value of the top 10 hypotheses is 182.366 and the fixed cost value is 75.791. Thus, a difference of 106.575 bits between fixed cost and null cost consigns to a meaningful pharmacophore model. Moreover, the total cost of the generated hypothesis should be closer to the fixed cost. All ten generated hypotheses scored a total cost closer to the fixed cost which leads to a good model. Statistically significant factors which include cost values, correlation coefficients (r), pharmacophore features, and root mean square deviations (RMSDs) of all 10 hypotheses are listed in Table 1 . The configuration cost enumerates the entropy of the hypothetical space and its value should not exceed a maximum value of 17 for a significant pharmacophore model [59, 60] . The configuration cost value of 16.601 was obtained for this pharmacophore generation calculation. Seven of the 10 hypotheses were made of five pharmacophoric features while another three had shown four features. The HY-AR was the common feature among all hypotheses. Nine of the 10 hypotheses had Hydrogen-bond acceptor (HBA), three hypotheses had ring aromatic (RA) while only one hypothesis was made of hydrogen bond donor (HBD). Hypo1 consists of two HBA and three HY-AR features and scored the better correlation and cost difference values. The RMSD value indicates the quality of "prediction" for the training set. The RMSD of all ten hypotheses ranged from 1.176 to 1.421 Å while the Hypo1 showed the lowest RMSD value of 1.176 Å. The correlation coefficient for the Hypo1, 0.942, represents a good correlation by linear regression of the geometric fit index. All these results construe that Hypo1 is the best ranking pharmacophore model among other hypotheses ( Figure 4 ). On the basis of the activity, compounds belonging to training and test sets were categorized into activity scales: most active (++++, IC 50 (inhibitory concentration) < 20 nM); moderately active (+++, ≥20 IC 50 < 200 nM); less active (++, ≥200 IC 50 < 2000 nM); inactive (+, IC 50 ≥ 2000 nM). Activities of all compounds were estimated based on the best ranking pharmacophore model, Hypo1. The experimental and estimated activity values for the 20 training set compounds based on Hypo1 are listed in Table 2 . Analysis of the activity prediction of training set compounds revealed that all the most active compounds were predicted in the same scale, whereas only one moderately active compound was estimated as less active and three inactive compounds were estimated as less active compounds among the 20 compounds of training set. The estimated activity values of most and least active compounds of the training set based on Hypo1 were 0.27 and 4800 nM, respectively, which are very close to that of their experimental activity values (0.46 and 5900 nM). This result revealed that the structural characteristics which can explain the difference in their biological activities are present in Hypo1 ( Figure 5a ). The most active compound 1 could map all the features of the best pharmacophore model, Hypo1, with a fit value of 9.04. The carbonyl oxygen atoms attached with the piperazine ring and azetidinone moiety were mapped onto the two HBA features. All three phenyl rings present in this most active compound mapped over three HY-AR features. The least active compound 20 in the training set maps Hypo1 with a fit value of 4.79 missed two HY-AR features as compared to compound 1. Carbonyl group of imidazolidine-dione and the only carboxyl group of this least active compound mapped both the HBA features whereas the phenyl ring attached to the imidazolidine-dione mapped over one of the HY-AR features ( Figure 5b ). The validation of suggested pharmacohore model, Hypo1, was performed by two different validation methods, namely, test set prediction and Fischer randomization methods. A test set containing 97 compounds, representing diverse activity classes and different functional groups, is used in this validation process. These test compounds were imported into the DS and diverse conformers were built in the same manner as for training set compounds. The estimated activities of these test set compounds were calculated based on the geometric fit of these compounds over Hypo1. Analyses of the estimated activities of test set compounds demonstrated remarkable results. From the 97 test set compounds, 94 compounds showed error values less than 5 which is hardly different from the experimental and estimated activity values (Table 3) . Eight out of nine of the most active compounds were estimated in the same activity scale, whereas the ninth compound was predicted as moderately active. Seventeen out of 26 moderately active compounds were estimated in the same scale, whereas the remaining nine were estimated as less active compounds. All the 40 less active compounds were estimated in the less active scale. Furthermore, only three of the 22 inactive compounds were predicted as less active compounds. Thus, the ability of Hypo1 to forecast the activity of test set compounds was very impressive and outstanding. A correlation value of 0.928 was achieved between experimental and estimated activities of test set compounds. A correlation plot showing the correlation between the experimental and estimated activity values of training and test set compounds was generated and displayed in Figure 6 . Another validation method based on Fischer randomization was also performed on the training set compounds to verify the quality of Hypo1. In this validation process, a confidence level of 95% was selected and thus 19 spreadsheets (Table 4 ) were generated. The data obtained from this validation method did not produce any better statistical values compared with that of Hypo1. Out of the 19 runs, only three had a correlation value between 0.90 and 0.92 which was comparatively less than the correlation value of Hypo1. The total cost values of all randomized models and RMSD values were higher than Hypo1, which is not appropriate for a good pharmacophore model. Therefore, this validation test also endows the Hypo1 with a high level of assurance. The suggested pharmacophore model Hypo1 developed so far divulges a fairly accurate idea of the required molecular features for a new lead. Therefore, Hypo1 was applied as a search query to retrieve molecules with novel and desired attributes from chemical databases (Maybridge and Chembridge). A total of 2202 hit compounds, 1478 compounds from Maybridge and 724 compounds from Chembridge, respectively, were obtained. Molecular properties were calculated for all hit compounds retrieved from databases. The 181 hit compounds (124 from Maybridge and 57 from Chembridge database, respectively) with an estimated activity value closer to the most active compound in the training set were selected for further evaluation. These hits were further filtered by using Lipinsiki's rule of five which evaluates drug-likeness, or determines if a chemical compound with a certain pharmacological or biological activity has properties that would make it feasible to be an orally active drug in humans. The 49 compounds of Chembridge database and 23 compounds from Maybridge database have satisfied the requirements of Lipinsiki's rule of five for a drug-like compound. Thus, these 72 hit compounds that satisfied the Lipinsiki's rule of five from a total of 181 hits were subjected to molecular docking. All of the 20 training set compounds along with the 72 database hits retrieved from the database screening process were docked into the protein active site using the GOLD (Genetic Optimization for Ligand Docking) docking program. GOLD fitness score which differentiates molecules on account of their interacting pattern is calculated for all molecules. The most active compound of training set (compound 1) scored a docking score of 66.6 and exhibited various hydrogen-bonding interactions with the key active site residues (Figure 7a ). Moreover, two of the carbonyl oxygen atoms near the middle ring that mapped on the HBA features of "Hypo1" showed hydrogen-bonding interactions with Gly193 and Ser195 residues of the active site. Previous studies of chymase have also divulged the importance of Gly193 and Ser195 as key amino acids in active site region of the enzyme [9, 13] . Along with diverse hydrogen-bonding contacts, the phenyl group of compound 1, which was mapped on the HY-AR feature of "Hypo1" showed π···σ interactions with the aromatic ring of residue F191. Moreover, compound 1 also showed hydrophobic interactions with Y215 and L99 amino acids. Several hit compounds obtained from database screening process also showed high GOLD fitness scores and formed interactions with the active site residues. The hit compounds that showed a fitness score of more than 66 were selected as final hits for further evaluation process. Intriguingly, all the final three compounds were obtained from Maybridge database and none from the Chembridge database. Compound HTS12673 which showed an estimated activity value of 6.716 nM has scored a GOLD fitness score of 78.73. It has also exhibited key interactions with the important amino acids like Gly193, Ser195, Y215, and H57 at the active site of the enzyme (Figure 7b ). The phenyl part of anisole ring and pyridine ring that mapped over the HY-AR features of "Hypo1" instigated the improved binding of this compound through better hydrophobic interactions. Compound BTB02076, which was also retrieved from the Maybridge database, with an estimated activity value of 8.605 nM has shown a GOLD fitness score of 72.40. This compound has formed various close contacts that lead to the important ligand-enzyme interaction such as hydrogen bonding interactions with Gly193, Ser195 and hydrophobic interactions with Phe191 amino acid in the active site of the enzyme (Figure 7c) . Moreover, important π···π interactions between the fused ring system of BTB02076 and the side chain imidazole ring of His57 amino acid were also revealed. Furthermore, it also showed hydrophobic interactions with Y125 and L99 amino acid residues of protein through the hydrophobic groups mapped over HY-AR features of Hypo1. Third hit, JFD00311, with the estimated activity value of 4.661 nM and GOLD fitness score of 74.51 has formed hydrogen bond network with the active site residues Gly193, and Ser195 (Figure 7d ). The benzene rings and oxygen atoms of the benzenesulfonic acid moieties in this hit compound that overlaid the HY-AR and HBA features of "Hypo1", respectively, enabled considerable hydrophobic and polar interactions with the important amino acids in the active site. The mapping of these top three database final hits on Hypo1 and their 2D molecular structures are depicted in Figures 8 and 9 , respectively. All three hit compounds have mapped the entire features of the best pharmacophore model, Hypo1. Thus, in the design of potent inhibitors of chymase, compounds HTS12673, BTB02076, and JFD00311 which showed important results with respect to all properties such as estimated activity, calculated drug-like properties and better GOLD fitness scores can be proposed as potential leads. Novelty search using SciFinder Scholar and PubChem compound search has also ascertained that these hits were not reported earlier for chymase inhibition. The electrostatic features impacting the inhibitory effect of chymase inhibitors have been investigated aiming at providing useful information for understanding the structure inhibition relationships of chymase inhibitors. Structures of the most and least active compounds of the training set are optimized along with the three final database hit compounds at B3LYP/6-31G* level. Statistically significant factors such as HOMO, LUMO, and MESP, for all compounds are calculated. According to Fukui's frontier orbital approximation, the frontier orbitals HOMO and LUMO of a chemical species are very important in defining its reactivity. Fukui first recognized the importance of frontier orbitals as principal factors governing the ease of chemical reactions and the stereoselective path while Parr and Yang demonstrated that most frontier theories can be rationalized from DFT. When the whole dataset of molecules was taken into account, an apparent trend of inhibitory activity (IC 50 ) data with an increase in HOMO energy was observed ( Figure 10 ). For all compounds, HOMO energy ranges between −5.619 and −6.415 eV. High value of E HOMO is likely to indicate a tendency of the molecule to donate electrons to appropriate acceptor molecule of low empty molecular orbital energy. The correlation of HOMO energies with IC 50 data indicates that the HOMO of the inhibitor may transfer its electrons to less energy, LUMO, of some amino residues in the active site of chymase. The calculations show that compounds 1 and 20 have shown the highest (−5.873 eV) and lowest (−6.415 eV) HOMO level energies respectively. This trend is in good agreement with the experimental observations suggesting that compounds 1 and 20 have exhibited the highest (0.46 nM) and lowest (5900 nM) inhibitory profile, respectively, in all investigated chymase inhibitors. While BTB (BTB02076) has shown highest (−5.619 eV) HOMO level energy among hit compounds even higher than HOMO energy level of compound 1, the other two hit compounds also showed higher E HOMO than the least active compound of the data set. In a previous study, a high HOMO energy level also played an important part in activity of the most active dual and selective LOX inhibitors [61] . Moreover, a clear trend between the inhibitory activity (IC 50 ) data and LUMO energy of all compounds was also revealed. For all compounds, LUMO energy ranged between −0.631 and −2.275 eV. Compound 1 and BTB showed highest LUMO level energies; and least active compounds 19 and 20 demonstrated LUMO with lowest energies. HOMO and LUMO sites are plotted onto the molecular surface of most active (1) and least active (20) compounds of the data set along with the two hit (BTB, HTS) compounds ( Figure 11 ). Most often, the heteroaromatic rings, which contain the heteroatoms such as nitrogen and oxygen, are the regions in all these compounds that can act as electron donors or acceptors to the active site of the chymase. Experimental study also deduced that introduction of heteroatoms to the inhibitor compound enhanced its stability in human plasma (20) . For instance, the placement of an ethoxy group in compound 2 instigated its stability. Electron donor rings can be identified as those with the greatest electron density from the HOMO. In the case of compound 1, HOMO is scattered over the 4-methylpiperazine moiety together with the carbonyl group and LUMO is spread over the region 2-hydroxyl-4-oxoazetidine containing heteroatoms like oxygen and nitrogen. Docking results also showed that this region of compound 1 is involved in important interactions with the key residues of protein. For compound 20, HOMO is composed of aniline ring and LUMO spreads over sulfonyl and benzoic acid groups. LUMO plot over methylbenzenesulfonamide group in hit compounds BTB showed hydrogen bonding interactions with important amino acids Gly193 and Ser195 at the active site of the enzyme. Whereas the HOMO plot is scattered on 2-methoxyphenol group and dihydroquinazolin moiety, the six membered ring part of dihydroquinazolin group is involved in important π···π interactions with the side chain imidazole ring of His57 amino acid. For HTS hit compound, HOMO and LUMO are composed of methoxybenzene, benzoindazole moieties, and oxadiazole substituted pyridine moiety, respectively. Overlay of HTS on "Hypo1" and its docking with the protein also speculated the involvement of these groups in key interactions with the active site of protein. The effect of the orbital energies on the inhibition activities can be associated with the charge transfer, π···π, or π···σ stacking between inhibitors and aromatic amino acid residues in the binding site of chymase. The result of molecular docking studies on chymase inhibitors also proved the presence of such kind of interactions. Electrostatic potential is widely used in characterizing molecules, especially for biomolecules, and takes special effect in the biomolecular recognition and in the prediction of the functional sites [62] . Nam et al. reported their discovery that electrostatic interactions accounted for the majority of the rate acceleration in the mechanism of RNA transphosphorylation in solutions catalyzed by the hairpin ribozyme [63] . Daga and Doerksen have stated the binding mode and the role of stereoelectronic properties in binding of spiroquinazolinones showing phosphodiesterase 7 (PDE7) inhibitory activities [64] . Recently, the electrostatic funnel illuminated from three-dimensional mapping of the electrostatic potential was reported by Dehez et al., driving the diphosphate nucleotide rapidly toward the bottom of the internal cavity of membrane-protein mitochondrial ADP/ATP carrier by forming a privileged passageway [65] . Considering these discoveries comprehensively, we supposed that the electrostatic potential of the inhibitor also played a significant role in the binding and interaction with chymase together with orbital energy and consequently influenced the inhibition effect. The 3D isosurface maps of MESP were interpolated on the electron density surfaces of constant electron charge density (0.0004 e/au 3 ). As is well known, the electrostatic potential is defined as the interacted energy of a positively remote charge point with the nuclei and the electrons of a molecule. The 3D plots of electron density (ED) and the MESP for compounds 1, 20, BTB and HTS are shown in Figure 12 . The red and the blue color represent the electronegative and electropositive potentials whereas the green represents a potential halfway between the two extremes. The coloring area of the surface represents the overall molecular charge distribution with the electrostatic potential. As for the compounds in this study, the electronegative potential (MESP min ) was coded with red on the MESP maps in a range from 202.16 to 152.27 kcal/mol indicating a strongest attraction while the interpolated blue map represents the electropositive potential (MESP max ) of a strongest repulsion varying from 15.68 to 42.67 kcal/mol. The predominance of green region in the MESP surfaces corresponds to a potential halfway between the two extremes that are indicated in red and blue colors, respectively. MESP plotted onto constant electron density surface for most active compound 1 showed the most electronegative potential region (red color) over the oxygen atom of the carbonyl group near the piperazine moiety. However, in the case of the least active compound 20, most negative potentials due to sulfonyl and carbonyl oxygen atoms are missing. For hit compounds, appearance of localized negative potential regions located at the oxygen atoms of the carbonyl and sulfonyl groups and nitrogen of the pyridine ring are consistent with the docking results which recognized this region as hydrogen bond acceptor. Moreover, one more prominent localized negative charged region protruding over the oxadiazole group was oriented adjacent to Gly193, to be recognized as a hydrogen bond acceptor. The strong electrostatic interaction of the negative potential with key residues Gly193 and Ser195, namely the formation of the hydrogen bond, will enhance the inhibition effect substantially together with the orbital interaction through the exchange of energy. The blue electropositive maps of these compounds were mainly distributed over the methyl group. The hydrogen atoms attached to the six-membered rings also bear the maximum brunt of positive charge (blue region). Due to the accumulation of positive potential, these moieties exhibited π···π and π···σ interactions with the aromatic residues of active site. These molecular electrostatic potential features are also in concert with the key chemical features (HBA and HY-AR) of pharmacophore model (Hypo1) which was successfully employed as a 3D structural query for virtual screening of databases for the identification of new potent chymase inhibitors. Thus electrostatic potential of the inhibitors can play a significant role in the binding and interaction with chymase together with orbital energies, and consequently influence the inhibition effect. A set of 117 structurally distinct compounds reported as chymase inhibitors with their diverse experimentally known inhibitory activity (IC 50 ) data was compiled from the literature such as life science journals [14] [15] [16] [17] [18] [19] [20] 55, [66] [67] [68] . All of the inhibitory activities were obtained using the same biological assay method [14] . To form a training set, 20 compounds with distinctive structural motif and wide activity range (0.46 to 5900 nM) were selected. For all compounds in the training set, energy minimization process was performed with CHARMM forcefield. Poling algorithm was applied to generate a maximum of 255 diverse conformations with the energy threshold of 20 kcal·mol −1 above the calculated energy minimum for every compound in the dataset. These conformers were generated using Diverse Conformer Generation protocol running with Best/Flexible conformer generation option as available in Accelrys Discovery Studio v2.5 (DS), Accelrys, San Diego, CA, USA. This method ensures the best coverage of conformational space by performing a more rigorous energy minimization in both torsional and cartesian space by using poling algorithm. All the 20 training set compounds associated with their conformations were submitted to the HypoGen module of DS. The HypoGen algorithm implemented for the pharmacophore hypothesis generation process is executed in three phases, namely, constructive, subtractive, and optimization phases. In constructive phase, identification of features common to the most active compounds takes place whereas all pharmacophoric features that are also present in the least active compounds are removed in subtractive phase. Finally, in the optimization phase, the hypothesis score is improved by regression parameters which are used for the estimation of the activity value of each training set compound. The relationship between the geometric fit value and activity value is utilized for this computation. Pharmacophore hypotheses showing best correlation in the 3D arrangement of features in a given training set compounds with the corresponding pharmacological activities are formed and ranked. Several structure activity relationship (SAR) pharmacophore models were derived from training set compounds using HypoGen module of DS. In this study, the top 10 hypotheses which were returned by the hypotheses generation process with significant statistical parameters were selected for further calculations. The generated quantitative pharmacophore model was validated to find out whether it is competent enough to identify the active structures and estimate their activity values precisely. This validation process was performed based on test set prediction and Fischer randomization methods. In developing statistical models, the following three cross-validation methods are often used to examine a model or predictor for its effectiveness in practical application: independent dataset test, subsampling test, and jackknife test [69] . However, of the three test methods, the jackknife test is deemed the most objective [29] . The reasons are as follows. (i) For the independent dataset test, although all the samples used to test the model or predictor are outside the training dataset used to train it so as to exclude the "memory" effect or bias, the way of how to select the independent samples to test the model or predictor could be quite arbitrary unless the number of independent proteins is sufficiently large. This kind of arbitrariness might result in completely different conclusions. For instance, a model or predictor achieving a higher success rate than the other model or predictor for a given independent testing dataset might fail to keep so when tested by another independent testing dataset [69] ; (ii) For the subsampling test, the concrete procedure usually used in literatures is the 5-fold, 7-fold or 10-fold cross-validation. The problem with this kind of subsampling test is that the number of possible selections in dividing a benchmark dataset is an astronomical figure even for a very simple dataset, as elucidated demonstrated by Equations 28-30 in [70] . Therefore, in any actual subsampling cross-validation tests, only an extremely small fraction of the possible selections are taken into account. Since different selections will always lead to different results even for a same benchmark dataset and a same model or predictor, the subsampling test cannot avoid the arbitrariness either. A test method unable to yield a unique outcome cannot be deemed as a good one; (iii) In the jackknife test, all the samples in the benchmark dataset will be singled out one-by-one and tested by the model or predictor trained by the remaining samples. During the process of jackknifing, both the training dataset and testing dataset are actually open, and each sample will be in turn moved between the two. The jackknife test can exclude the "memory" effect. Also, the arbitrariness problem as mentioned above for the independent dataset test and subsampling test can be avoided because the outcome obtained by the jackknife cross-validation is always unique for a given benchmark dataset. Accordingly, the jackknife test has been increasingly and widely used by those investigators with strong math background to examine the quality of various predictors (see e.g., [30, [71] [72] [73] [74] [75] ). However, to reduce the computational time, we adopted the independent testing dataset cross-validation in this study as done by many investigators with SVM as the prediction engine. A test set comprising 97 compounds with experimentally known chymase inhibitory activity values was used in test set prediction method. Ligand Pharmacophore Mapping protocol running with BEST/Flexible conformation generation option was used to map the test set compounds. Fischer randomization method as available in DS was applied on training set compounds to prove that the generated pharmacophore model was not obtained by chance. A pharmacophore is only useful as a predictive model in finding novel, potential leads suitable for further development only if it is able to detect the compounds with known inhibitory activity. In order to identify new potential lead compounds, the selected pharmacophore model was subsequently used as 3D structural search query to screen the Maybridge and Chembridge chemical databases consisting of 60,000 and 50,000 of structurally assorted compounds, respectively. All queries were performed using Ligand Pharmacophore Mapping protocol running with Best/Flexible search method in DS. To be retrieved as a hit, a molecule must fit all the features of the pharmacophore hypothesis. The hits obtained through database screening were further filtered using Lipinsiki's rule of five in order to carry only drug-like compounds in further studies. Computational docking operation is a useful vehicle for investigating the interaction of a protein receptor with its ligand and revealing their binding mechanism as demonstrated by a series of studies [18, 46, [76] [77] [78] [79] [80] [81] [82] [83] [84] . Docking plays a significant role in predicting binding orientation and affinity of small molecule drug candidates to their protein targets with known 3D structures [85, 86] . Hence, docking serves as an important tool in the rational computer-assisted drug design [87, 88] . GOLD 4.1 (Genetic Optimization for Ligand Docking) from Cambridge Crystallographic Data center, UK uses a genetic algorithm for docking ligands into protein binding sites to explore the full range of ligand conformational flexibility with partial flexibility of protein [89] . In this study, it has been utilized for the docking of training set compounds along with the new hits retrieved from chemical databases. Protein coordinates from the crystal structure of chymase co-crystallized with β-ketophosphonate (PDB ID: 1T31), determined at a resolution of 1.9 Å were used to define the active site [9] . All the water molecules present in the protein were removed and hydrogen atoms were added. The active site was defined with a 10 Å radius around the ligand present in the crystal structure. At the end of the computation, the 10 top-scoring conformations of every ligand were saved. Early termination option was applied to pass over the genetic optimization calculation when any five conformations of a particular compound were envisaged within an RMS deviation value of 1.5 Å. The GOLD fitness score is calculated from the contributions of hydrogen bond and van der Waals interactions between the protein and ligand, intramolecular hydrogen bonds and strains of the ligand. The protein-ligand interactions were scrutinized by DS. As far as computational technique is considered, many practices have ascertained that DFT, which takes into account the exchange and correlation effects effectively, is most likely one of the best methods to study medium-size or larger molecular systems and appropriate for QSAR study, with exhibiting excellent performance than semiempirical method or some other ab initio methods. Complete geometry optimization for data set compounds was carried out using DFT with Becke's three-parameter exchange potential and the Lee-Yang-Parr correlation functional (B3LYP), using basis set 6-31G* level [90] . A useful kind of net atomic charges, called electrostatic potential (ESP)-fitting charges, were derived from the DFT calculated molecular electrostatic potential distribution using CHelpG method, which produces charges fit to the electrostatic potential at points selected. Vibrational frequencies were computed at the same B3LYP/6-31G* level to characterize the stationary points on the corresponding potential energy surfaces. All calculations were performed using the Gaussian 03 suite of programs. Based on structural diversity and wide biological activity range, four chymase inhibitors including most and least active compounds, were selected from the training set. While, three final hits BTB02076, HTS12673, JFD00311 retrieved from Maybridge database by the selected pharmacophore model, which showed important results with respect to all properties like molecular interactions with the active site components, estimated activity, calculated drug-like properties, and high GOLD fitness score, were also selected. Thus, data set employed for DFT study consisted of seven compounds. Various quantum-chemical descriptors such as LUMO, HOMO, and locations of molecular electrostatic potentials (MESP), were computed. The mapping of the electrostatic potential is an established technique for investigation of biologically active compounds because it plays a key role in the initial steps of ligand-receptor interactions. The formatted checkpoint files of the compounds generated by the geometric optimization computation were used as input for CUBEGEN program interfaced with Gaussian 03 program to compute the MESP. The MESP isopotential surfaces was produced and superimposed onto the total electron density surface (0.0004 e/au 3 ). The electrostatic potential of the whole molecule is finally obtained by superimposing the electrostatic potentials upon the total electron density surface of the compound. Since user-friendly and publicly accessible web-servers represent the future direction for developing practically more useful models or predictors [91] , we shall make efforts in our future work to provide a web-server for the method presented in this study. Combining various theoretical methods like pharmacophore modeling, database screening, molecular docking and DFT calculations, an investigation for identification of novel chymase inhibitors and to specify the key factors crucial for the binding and interaction between chymase and inhibitors was performed. The highly correlating (r = 0.942) pharmacophore model (Hypo1) with two hydrogen bond acceptors, and three hydrophobic aromatic features was generated. After successfully validating "Hypo1" using test set and Fischer randomization methods, it was further used in database screening. Hit compounds were subjected to various drug-like filtrations and molecular docking studies. Finally, three structurally diverse compounds with high estimated activity and strong molecular interactions with key active site amino acids were identified. Furthermore, a DFT study, which articulated the influence of the electrostatic features of compounds on their inhibitory activity well, was performed. Analysis of orbital energies and plots of MESP has shown very clear trends between electronic properties and inhibitory activity (IC 50 ) data. An increasing trend was observed between IC 50 and HOMO energy values. The molecular electrostatic potential features were also consistent with the key chemical features of "Hypo1" thus successfully validating "Hypo1" by the DFT method. Therefore, the results of this study will be helpful, not only in the development of new potent hits for chymase, but also in providing a better understanding of the interaction between the chymase and inhibitors. This will in turn assist in the rational design of novel potent enzyme inhibitors.

Perspectives on Immunoglobulins in Colostrum and Milk

Immunoglobulins form an important component of the immunological activity found in milk and colostrum. They are central to the immunological link that occurs when the mother transfers passive immunity to the offspring. The mechanism of transfer varies among mammalian species. Cattle provide a readily available immune rich colostrum and milk in large quantities, making those secretions important potential sources of immune products that may benefit humans. Immune milk is a term used to describe a range of products of the bovine mammary gland that have been tested against several human diseases. The use of colostrum or milk as a source of immunoglobulins, whether intended for the neonate of the species producing the secretion or for a different species, can be viewed in the context of the types of immunoglobulins in the secretion, the mechanisms by which the immunoglobulins are secreted, and the mechanisms by which the neonate or adult consuming the milk then gains immunological benefit. The stability of immunoglobulins as they undergo processing in the milk, or undergo digestion in the intestine, is an additional consideration for evaluating the value of milk immunoglobulins. This review summarizes the fundamental knowledge of immunoglobulins found in colostrum, milk, and immune milk.

The topic of immunoglobulins in milk immediately brings to mind the relationship between mother's milk, transfer of passive immunity from mother to neonate, and the immature immune system of the neonate. Research in this field dates back to the late nineteenth century, however for many centuries herdsmen have capitalized on the linkage between maternal immune status and the immunological protection and development of the neonate [1, 2] . Immunoglobulins in mammary secretions come from several sources and represent a history of the antigen exposure of the mother and the response of her immune system. Immunoglobulins are transported through the mammary epithelial cells by receptor-mediated mechanisms and transferred out of the mammary gland by milk ejection during suckling. The immunoglobulins then enter the environment of the gastrointestinal tract of the neonate. Although that environment is primarily geared toward digestion to gain nutritional benefit, the immunoglobulins remain sufficiently stable to provide protective benefits for the neonate, either through uptake into the vascular system in the newborn of some species or through immunological function in the gastrointestinal tract. The immunoglobulins found in milk and the transfer of passive immunity from mother to neonate have been reviewed by many authors, with a partial listing referenced here [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] . In addition to the importance of homologous transfer of passive immunity between mother and neonate, there is considerable interest in the potential for heterologous transfer of passive immunity, such as immunoglobulins obtained from one species and utilized for passive immunity in another species. The ability to manipulate the immunological status of animals through vaccination against diseases that affect humans and the opportunity to harvest those immunoglobulins in the form of colostrum or milk has long been recognized [19, 20] , and continues to be a topic of interest in both animal science and human medicine [13, 16, 17, [21] [22] [23] . This review begins with a summary of some of the research on what has been termed -immune milk‖ and then discusses various aspects of immunoglobulins in mammary secretions (structure, function, concentration, sources, transport, species differences, and roles of immunoglobulins). Finally, traits related to stability and processing methods for collecting milk immunoglobulins are reviewed. One intriguing application of our knowledge about bovine colostral and milk immunoglobulins comes through the opportunity to provide passive immunity against diseases in other species, especially in humans. The ability to direct the cow's immune system to produce antigen-specific antibodies that are secreted in colostrum and milk and may be used to provide protection against a specific disease continues to be an area of interest. For example, the widespread consumption of immune milk from cows inoculated against diseases such as avian influenza, SARS, and other human respiratory diseases, has been suggested as a potential means of slowing outbreaks of the disease before they reach epidemic levels [24] . A number of reviews have summarized and evaluated early attempts to develop and test the use of immune milk products to provide passive immune protection [21] [22] [23] [25] [26] [27] [28] [29] [30] [31] . Several immune milk products are available commercially [13, 17, 22, 23, 32] . Safety issues associated with use of bovine immune milk products for human use have been discussed by others [23, [33] [34] [35] . The discussion below provides some examples of immune milk products and their use against some animal and human diseases (sections 2.2-2.7). Secretion of antibodies in breast milk from naturally immunized mothers can provide protection against enteric and other diseases in children [11] . For example, elevated concentrations of antibodies specific for enteric pathogens, such as Vibrio cholerae, in the mother's breast milk do not prevent colonization with the bacterium in the nursing child, but do seem to protect the infected child from developing diarrhea [36] . Breast feeding is associated with a reduced incidence of Campylobacter diarrhea in young children compared with children that do not breast feed [37] . In those children that are breast fed and do develop diarrhea, the human milk consumed may not contain IgA antibodies specific for the common antigen of Campylobacter [37] , suggesting a degree of antigen specificity contained in the breast milk. The idea of immunizing the pregnant animal with the intent of controlling neonatal morbidity and mortality is well established [38] . Vaccination or natural immunization of the pregnant cow, ewe or sow against enterotoxigenic Escherichia coli [38] [39] [40] or intestinal viruses [41, 42] , can provide a degree of protection for the newborn. As an example, while only limited protection against viral challenge occurred in calves fed once shortly after birth with a pooled colostrum from cows immunized against bovine rotavirus, a shorter duration of diarrhea was observed [43] . On the other hand, calves fed milk supplements with low levels of a similar immune colostrum at each feeding for two weeks did have reduced virus sheading and reduced incidence of diarrhea [44] . In primates, immunization of pregnant baboons with a rhesus rotavirus vaccine increased milk immunoglobulin and virus neutralizing titre [45] . Prenatal immunization of pregnant women with a single dose of meningococcal vaccine not only increased antigen-specific IgG antibody in the infant's serum during the initial 2-3 months after birth, but antigen-specific IgA in milk continued to be elevated at least up to 6 months [46]. As discussed in section 4, IgG transfer to the offspring in humans occurs during late pregnancy and provides the initial systemic source of that immunoglobulin. Infants consuming breast milk will primarily be consuming secretory IgA (section 4), which has significant protective activity in the intestine, as discussed in section 5.2. The above examples of homologous transfer of passive immunity set the stage for considering the opportunities for heterologous passive transfer. Immune milk products generally are some form of protein product derived from the colostrum and/or milk of dairy cattle. The cows typically are hyperimmunized against one or more antigens representing pathogens of bacterial or viral origin. Crude preparations of the immunoglobulin from colostrum or milk may range from essentially no alteration of the immunoglobulin concentration in the product to partial immunoglobulin isolation or concentration in a whey protein concentrate. The primary immunoglobulin in cow colostrum and milk is IgG, whereas the primary immunoglobulin in human milk is IgA [1] . Nevertheless, bovine IgG from colostrum or milk can be effective as a means of providing passive immunity to protect animals and humans from disease. The use of bovine colostral immunoglobulin preparations from immunized cows for disease protection of the neonate of other species has been demonstrated in swine [47] , and experimental animal models such as mice [48, 49] . There also are a number of examples of the use of bovine immune milk products in the treatment or prevention of human disease, especially in cases where the pathogen acts by way of the gastrointestinal tract. When considering these studies, it should always be kept in mind that the colostrum or milk preparations potentially contain other immune modulating substances than immunoglobulins, as discussed briefly below (section 6.3). The concept of using immune milk derived from hyperimmunized cows for treatment of human disease can be traced back to the 1950s and earlier [19, 20] . Some of the early efforts in this field involved using immune milk products for treatment of rheumatoid arthritis and hay fever [19] . Immune milk preparations produced from milk from cows immunized with a heat-killed, lyophilized mixture of bacteria found to reside in the human gastrointestinal tract has been studied for the prevention and treatment of rheumatoid arthritis, high blood cholesterol, high blood pressure, and oral submucous fibrosis [50] [51] [52] [53] . On the other hand, most studies on the use of immune milk have examined the potential of immune milk for prevention and treatment of infectious diseases, particularly gastrointestinal disease. Even milk that does not come form hyperimmunized cows may in some sense be regarded as immune milk. Bovine anti-human rotavirus IgG1 antibodies have been found in raw and pasteurized milk from cows that had not been specifically immunized against that virus [54] . Milk from non-immunized cows also has been found to contain measurable antigen-binding activity against several human pathogenic bacteria [55] . Several authors have tested the efficacy of immunoglobulin preparations with antibody activity against human rotavirus as a means of providing passive immunity to children. For example, children consuming a defatted colostrum preparation from cows immunized against a strain of human rotavirus had no improvement of symptoms when the infection was established (patients admitted to a hospital with rotavirus infection), however the preparation was effective in limiting diarrhea in children when consumed prior to the infection [56, 57] . In another study, cessation of excretion of rotavirus in the stool of infants with acute rotavirus gastroenteritis was correlated with the presence of neutralizing activity in the stool after ingestion of a bovine whey protein concentrate from rotavirus-hyperimmunized cows [58] , although there was not a significant decrease in duration of diarrhea in that study. Other studies have found that treatment of children with hyperimmune bovine colostrum from cows immunized with human rotavirus serotypes reduces the duration and severity of diarrhea due to rotavirus [59] , and can provide significant protection from rotavirus infection [60] . Enteropathogenic bacteria have also been the target for development of immune milk. Over 80% of childrens' stools became negative for the E. coli strains used to hyperimmunize the cows that provided the source of immunoglobulin in a bovine colostrum/milk immunoglobulin concentrate consumed by children for 10 days [61] . Interestingly, only one in nine children treated with the immunoglobulin concentrate, and having diarrhea that was associated with E. coli strains which were not used in the immunization of the cows, developed negative stools, underscoring the importance of the bacterial strain-specificity of the immune product. Consumption of a hyperimmune immunoglobulin concentrate with a high antibody titer against a lipopolysaccharide isolated from Shigella flexneri 2a also has been shown to provide protection against a challenge with the same strain [62] . However, no difference in diarrhea or other symptoms in children with stools positive for S. dysenteriae was found whether treated with bovine colostrum from cows immunized against S. dysenteriae or with colostrum from cows not hyperimmunized [63] . Enterotoxigenic E. coli also is commonly associated with traveler's diarrhea. Prophylaxis against this infection may be achieved by providing passive immunity with immune milk. A bovine whey protein concentrate from cows immunized with enterotoxigenic E. coli serotypes and consumed 3-times daily for seven days protected all of the adult volunteers from developing diarrhea after being challenged with an enterotoxigenic E. coli strain [64] . In contrast, 90% of the volunteers who received control immunoglobulin concentrate prior to challenge developed diarrhea after the E. coli challenge. Subsequent studies using IgG isolated from bovine colostrum from cows hyperimmunized against specific E. coli colonization factor antigens also have shown protective effects in volunteers challenged with colonization factor antigen-bearing enterotoxigenic E. coli [65] , however other studies by the same group did not demonstrate significant effects of similar milk immunoglobulin products [66] . Bovine colostrum concentrate preparations derived from cows that have not been hyperimmunized against specific antigens also may provide some benefit via passive immunization for some diseases. For example, a commercial product which is made from large standardized pools of colostrum collected from over 100 cows has been used to treat a number of diseases [22, 23] , including diarrhea caused by diarrheagenic E. coli [67] . Similar preparations from non-immunized cows may provide protection against bacterial toxins that are the cause of diarrhea in AIDS patients [68] . These studies, along with the above mentioned study comparing colostrum preparations from cows immunized against S. dysenteriae or non-immunized cows [63] , demonstrate that bovine colostrum contains significant antimicrobial properties as a result of natural exposure of the cows to antigens of pathogens that may afflict humans. Another example of a potential use for bovine immunoglobulin preparations to control bacterial populations comes from studies on dental caries formation [69] . The concept of prenatal immunization of the pregnant mother to protect the neonate against dental caries was demonstrated in rats [70] . In applications to humans, bovine whey preparations of colostrum from cows immunized with caries-inducing bacterial strains (Streptococcus mutans and Streptococcus sobrinus), and containing over 60% immunoglobulin of which 80% was IgG1, has been used in several studies evaluating its effect on caries-producing bacteria. The colostral whey preparation reduced adherence of Streptococcus mutans in vitro and caused aggregation of suspended bacteria [71] , as well as inhibited glucose uptake by the test organism [72, 73] . The whey preparation from hyperimmunized cows opsonized bacteria and enhanced in vitro phagocytosis of bacteria by human leukocytes [74] . Antibodies in the whey preparation remained functional when added to milk that had been treated via ultra-high temperature pasteurization or milk that was fermented to extend shelf-life [75] . Immune milk from cows hyperimmunized against seven Streptococcus mutans strains reduced the recoverable bacterium in plaque samples from volunteers within seven days of initiation of mouth rinsing with the whey concentrate product [76] . Mouth rinsing with immune milk collected from cows immunized with a fusion protein representing two of the major factors implicated in oral colonization by Streptococcus mutans inhibited recolonization of saliva and plaque by that organism [77, 78] . Immunodeficiency disorders often are associated with cryptosporidiosis, which can lead to chronic malabsorption and weight loss. In a case study of a child with congenital hypogammaglobulinemia, severe vomiting and diarrhea due to cryptosporidiosis, gastric infusion with hyperimmune bovine colostrum from cows immunized with cryptosporidium oocytes resolved the symptoms within a few days and oocyts were no longer found in stool samples after about eight days [79] . Similarly, in a child with AIDS who had severe diarrhea caused by cryptosporidiosis, administration of a commercial hyperimmune bovine colostrum preparation with anticryptosporidial activity improved the diarrhea and eliminated the parasite [80] . In the cases where immune milk is collected from cows immunized against one or more pathogens, the immunization regimen occurs during the prepartum period of the cow. To put this in perspective relative to the lactation cycle of a cow, a brief reminder of that cycle may be helpful. Depending on the management system used by a farm, most dairy cattle will have their first calf early in their third year, marking the start of their first lactation. The cow will be re-bred about two to three months into lactation. Pregnancy is approximately 280 days. At about 2 months before expected calving date, or approximately 10 months into lactation, milk removal is halted and the cow is given what is called a -dry‖ period. The mammary gland undergoes a process of involution during the early dry period where most residual milk components are broken down and resorbed [81] . The mammary gland begins a redevelopment phase several weeks prior to calving. Colostrum formation occurs in the days leading up to calving, coinciding with the early phase of lactogenesis (initiation of lactation). In the cow, lactogenesis begins shortly prior to calving and extends into the first few days postpartum. Colostrum collected at the first milking of the cow after calving represents the accumulation of colostral products during the days leading up to parturition, including immunoglobulins which are at their highest concentration in the first milking. Concentrations of immunoglobulins then decline rapidly in the subsequent several milkings [82] . One application for immunization of pregnant or lactating animals comes from the arena of mastitis control in cattle. Mastitis is the major disease in dairy cattle and most often is caused by intramammary infection [83, 84] . Vaccination of cattle against mastitis-causing pathogens has been an area of study for many years [85, 86] . Optimization of immunization schedules continues to be investigated [87] . Effective vaccines against mastitis-causing pathogens can increase antigen-specific immunoglobulins in the serum, which in turn can be increased in the mammary secretions. In the case of the J5 E. coli bacterin vaccine, the immunization also may be causing the mammary gland to become hyper-responsive to bacterial challenge [88] , reminding us that enhancement of antigen-specific antibodies in the milk is not the only mechanism by which the vaccine may be having its effect. Because the peripartum and early lactation periods are times of high susceptibility of the mammary gland to mastitis, many immunization schedules include prepartum immunizations during the dry period when milk is not removed and the mammary gland undergoes involution. It is also important to remember that cattle are generally immunosuppressed during the peripartum period [88, 89] , potentially compromising the impact of immunizations administered just before or just after calving. Coliform mastitis is one of the major types of mastitis in cattle [90] . The more successful vaccination protocols for mastitis control have been with the J5 E. coli bacterin vaccine which is administered initially either just before or at the time of drying off [87, [91] [92] [93] [94] [95] [96] . These typically are followed by additional vaccine doses approximately mid-dry period. Some protocols include an additional immunization within several days after calving [87, 91, 92, 94] , while others also continue immunizations into the first three months of lactation [87, 94] . Attempts to vaccinate against other mastitis-causing pathogens have been met with more limited success. Such vaccination protocols range from immunizations during the dry period [97] , to peak lactation [98] , and even late lactation [99] . Although most of the immunization protocols used in mastitis control administer the vaccine either intramuscularly or subcutaneously, intramammary immunization also can result in an increase in antigen-specific immunoglobulin in milk, as well as in the serum [100] [101] [102] [103] . A look across the immunization protocols used in studies to produce many immune milk products shows considerable variation, especially in the number and timing of immunizations. In those specifically collecting colostrum shortly after calving, multiple immunizations are administered during late pregnancy when the cow would be in the dry period [56] [57] [58] 60, 61, 64, 71, 72, 77, 79, [104] [105] [106] . Mammary secretions then are collected either only at first milking [79] , pooled from the first 4 to 6 milkings [56, 57, 72, 104] , pooled from the first 6 to 10 days after calving [58, 61, 64, 105] , or collected for longer periods into lactation [77] . Other studies have initiated immunizations during the late dry period and then continued vaccinating throughout lactation [50,51,76], or only vaccinating during lactation [107] . Many of these studies used intramuscular or subcutaneous immunization, although some also have incorporated intramammary [58, 79, 105] , or intravenous infusion [61] . Newer technologies for vaccine development and delivery may further enhance the production of immune milk products. Immunization protocols that expose animals to specific antigens may enhance humoral immune responses in the mammary gland, including peptide-based vaccines [108] , and DNA-based vaccines [109, 110] . Delivery of antigen to the animal can also be achieved with antigen encapsulated in biodegradable microspheres [111] , and with antigen-release devices [112] . Transgenic animals also have been used to produce antigens that then may be used to vaccinate animals against viral disease [113] . The immunoglobulins, or antibodies, found in colostrum or milk are the same as those found in the blood or mucosal secretions. They are a family of proteins with a range of protective bioactivities. Immunoglobulins are divided into several classes including IgM, IgA, IgG, IgE, and IgD [114] , and IgG, IgA and IgM are the major immunoglobulin classes in mammary secretions. IgM is the class that appears initially when an organism is exposed to an antigen for the first time (primary infection). IgM has a low specificity and hence a lower potency in defeating the infection. IgA is the major immunoglobulin class found in mucosal secretions and prevents mucosal infections by agglutinating microbes, whereas IgG is the primary immunoglobulin class found in bovine colostrum and milk. Several subclasses of IgG exist, with IgG1 and IgG2 being the major immunoglobulins in serum. All monomeric immunoglobulins have the same basic molecular structure, being composed of two identical heavy chains and two identical light chains, with a total molecular mass of approximately 160 kilodaltons (for details on immunoglobulin structure see [5, 14, 16] ). Both the heavy and light chains have constant regions and variable regions. Heavy and light chains are linked together by disulfide bonds, resulting in the classic Y-shape of the immunoglobulin molecule [114] . The number and location of the disulfide bonds is dependent on the class of immunoglobulin. Each immunoglobulin molecule has two antigen binding sites which comprise the antigen-binding fragment (Fab). The Fab includes the variable amino acid domain. At the other end of the molecule is the constant fragment (Fc) which has a constant amino acid sequence among molecules of the same subclass and which confers the identity of an immunoglobulin as a particular subclass. The Fc region of the molecule is the region that binds to Fc receptors on various cell types. In the case of polymeric immunoglobulins, including the polymeric forms of IgA and IgM that are found in milk, the monomeric forms of the immunoglobulins are linked together through the covalent interaction with a joining (J) chain [114, 115] . The result is a dimeric form of IgA and a pentameric form of IgM. Binding of these immunoglobulins to the J chain also results in them having several special features, including: a high valency of antigen-binding sites, allowing them to agglutinate bacteria; limited complement-activating activity, which allows them to act in a noninflammatory manner; and a high affinity for the polymeric immunoglobulin receptor (pIgR) that is responsible for transepithelial transport of IgA and IgM into mucosal secretions such as milk [116] . The pIgR and its relationship to the secretory component (SC) associated with secretory IgA and secretory IgM is discussed further below (section 5.2). The content of immunoglobulins in colostrum and milk is highly dependent on the animal species [1, 14] . The same holds for the relative proportion of the immunoglobulin classes. These species differences are adaptations to the reproductive strategies of the animals and the degree of maturation of the offspring at birth. Animal species may be divided into three classes [1] : (1) species where immunoglobulins are transferred mainly to the fetus via the placenta (humans and rabbits); (2) species where offspring are born agammaglobulinemic and immunoglobulin transmission occurs via mammary secretions (ungulates such as horses, pigs, cows, and goats); and (3) species where immunoglobulins are transferred both via placenta and mammary secretions (rats, mice and dogs). These adaptations have several consequences both for the composition of immunoglobulins in colostrum and milk, and for the role of colostrum. Indeed, for animals like rats, mice, dogs and ungulates, uptake of colostrum of adequate quality and sufficient quantity is important for the offspring to boost the systemic immune function in the short term, whereas colostrum consumption in the human infant provides more protection for the gastrointestinal tract (see section 6.3). This is reflected in a lower total immunoglobulin content in human colostrum as compared to colostrum from the other species ( Figure 1 ) [1, 3, 117] . Human colostrum has a low content of IgG (2%), and the IgG required to provide systemic immunity is transferred across the placenta before birth. In contrast, colostral IgG content in many other species is typically greater than 75% of the total immunoglobulin content (Figure 1 ). An additional consequence of different routes of immunoglobulin transmission relates to the changes in relative contents of immunoglobulins that occur in the transition from colostrum to milk within certain species (Figure 1 ). For example, the profile of immunoglobulins in human colostrum is similar to that found in milk, where the IgA level is high in both colostrum and milk (88-90% of total immunoglobulin). This is in contrast to the bovine mammary secretions where the high concentration of IgG in colostrum declines rapidly with successive milkings. For animals like rats, mice, dogs and ungulates, the role of colostrum and milk immunoglobulins is to provide immune protection both systemically and for the gastrointestinal tract, which is reflected in large changes in the profile of immunoglobulins during the transition from colostrum to mature milk ( Figure 1 ). Thus, for many species the proportion of IgA increases between colostrum and milk. [1] ; human and pig [3] ; and horse [117] . Immunoglobulins found in mammary secretions arise from systemic and local sources. In the case of IgG in milk, the major portion comes from the serum [14] . While IgG producing plasma cells may occur within the mammary tissue, their contribution to the IgG in colostrum is minor compared with the IgG derived from the serum. Although limited paracellular passage of immunoglobulins may occur during inflammation (mastitis), uptake and transport of immunoglobulin across the mammary epithelial barrier is thought to occur primarily through an Fc-receptor-mediated process [1, 7, [118] [119] [120] . Immunoglobulins are thought to bind to receptors at the basolateral surfaces of the mammary epithelial cell. These receptors are specific for the Fc portion of the immunoglobulin molecule. The receptor-bound immunoglobulin is internalized via an endocytic mechanism [121] , transported to the apical end of the cell and released into the alveolar lumen. Recent studies have shed additional light on the details of this process [122] . In the case of IgG, the receptor responsible for transcytosis of IgG into colostrum is referred to as FcRn, or the neonatal Fc receptor, because it was initially identified in the neonatal rodent intestine as the receptor responsible for the specific uptake of maternal IgG [123, 124] . The FcRn also has been implicated in the trans-placental transport of IgG in humans and other species [125] [126] [127] , which may involve an endocytic and transcytotic process [128] . Since its initial discovery, FcRn has been described in many tissues [122] . The receptor is a heterodimer composed of a membrane-bound α-chain similar to MHC class-1 molecules and a smaller MHC class-1 protein, β2-microglobulin [129] . Binding of IgG to FcRn is pH-dependent, with high affinity binding occurring at acidic pH, but only weak binding at neutral or basic pH [122] . This observation suggests that IgG taken up by the epithelial cells may bind to FcRn within an acidic environment in the endosomes. The precise mechanism of transport across the epithelial cell and release into the colostrum or milk remains to be demonstrated. The half life of IgG in serum is typically longer (1-3 weeks) than that for IgA or IgM (1-2 days), and the half-life of IgG2 is slightly longer than for IgG1 [122] . Evidence suggests that IgG2 has a higher affinity for FcRn than IgG1 [122] . In bovine colostrum, IgG1 is many fold greater in concentration than IgG2 [82] , although they are of approximately equal concentrations in serum. It may be that the majority of the IgG2 taken up by the mammary epithelial cell during colostrum formation is not passed on to the alveolar lumen, but rather is recycled back to the extracellular fluid. The FcRn is thought to have a major role in the recycling of IgG in various tissues in the body [130] [131] [132] . That is, IgG that potentially may be lost through various tissues is recycled by the respective cells by binding to FcRn and recycled back to the blood or lymph. This is supported by studies of overexpression of FcRn in transgenic mice where there is an extension of the half-life of serum IgG [133, 134] , as well as a boosting of the overall humoral immune response of the mice [135] . Localization of FcRn in bovine, sheep and water buffalo mammary tissue indicates that the receptor is homogeneously distributed throughout the epithelial cells prior to parturition, but primarily localized at the apical surface of the mammary epithelial cells after parturition [136] [137] [138] [139] . While this type of observation corroborates the conclusion that FcRn plays an important role in IgG transport during colostrum formation, at least in ruminant species, the precise meaning of this redistribution of FcRn staining in mammary cells remains to be determined. It is also interesting to note that the initial report of this distribution pattern in sheep mammary epithelium included the observation that the staining pattern became diffuse within the cells during mammary involution [136, 137] . Transport of IgG also may increase transiently in mammary secretions during involution in cattle [140] . Hormonal and local factors have been implicated in the control of immunoglobulin transport during colostrum formation [32] . Haplotypes of the FCGRT gene, coding for the MHC Class I α-chain of FcRn, are associated with serum concentrations of IgG in neonatal beef calves [141] and associated with IgG concentrations in colostrum of dairy cows [142] . Haplotypes of the β2-microglobulin gene (β2M) also are associated with serum IgG concentrations in newborn calves [143] . In estimating mass transfer of IgG1 into colostrum in dairy cattle, 10% of cows had mass transfer greater than one standard deviation above the mean, perhaps indicating a genetic or hormonal regulation of the variance of transport [144] . Clearly there is opportunity for genetic manipulation of IgG transport in the mammary gland to enhance the concentrations of immunoglobulins in colostrum and milk. However, it should be remembered that serum IgG concentrations in the periparturient cow are already decreased as a result of the extensive IgG transport into the colostrum [145] , and as indicated above, the cow is in an immunosuppressed state during the peripartum period [88, 89] . The other major classes of immunoglobulins transported into colostrum and milk are IgA and IgM. Immunoglobulin A is the major immunoglobulin in human colostrum and milk (Figure 1 ), however it is also present in milk of most other species. Colostrum and milk IgA and IgM are found in the form of secretory IgA, or sIgA, and sIgM. Much of these are produced by plasma cells in the mammary tissue. The plasma cells are part of the gut-associated lymphoid tissue (GALT), the largest immune organ of an organism, which includes the Peyer's patches, lymphoid and myeloid cells in the lamina propria and intraepithelial lymphocytes [146, 147] . Lymphocytes from the GALT system migrate to the mammary gland and provide a direct link between the antigen exposure response in the mother's mucosal immune system, especially via the enteric mucosal immune system, and the secretory immunoglobulin repertoire of the mammary gland [18] . This means that maternal colostrum and milk will contain antibodies specific for pathogens that may be encountered by the neonate's intestine and other mucosal tissues [10, 18, 148] , providing a rationale for the observations summarized above that bovine colostrum from nonimmunized cows also may afford passive immune protection against human pathogens [54, 55] . The immune connection between the GALT and the mammary gland is of particular interest with respect to human milk where the major immunoglobulin is sIgA, which accounts for one of the key factors underlying the importance of breast feeding [10] . The immune activation of GALT in the human infant is delayed, and the milk sIgA and sIgM provide the neonatal intestine a level of protection through their immune exclusion actions and their anti-inflammatory effects [10, 18] . Transepithelial transport of IgA and IgM across the mammary epithelial cells occurs via the polymeric immunoglobulin receptor (pIgR) which is responsible for binding dimeric IgA and pentameric IgM in mucosal tissues [149, 150] . The polymeric nature of IgA and IgM arises from their binding with the J-chain peptide [116] . Only IgA or IgM that contain the J chain have a high affinity for pIgR [116, 151, 152] . In fact, the J chain has been evolutionarily conserved within tetrapods to the point where human polymeric IgA can bind to the pIgR from the amphibian Xenopus laevis [152] . Polymeric IgA or IgM bound to pIgR is internalized and transported to the apical end of the mammary epithelial cell by an endocytic process. The pIgR molecule is cleaved to release a receptor fragment, called secretory component (SC), which remains bound to the immunoglobulin molecule [119, 149] . In the case of pIgR receptor sites that are not occupied by immunoglobulin, the secretory component is still cleaved from the membrane-bound portion of pIgR, resulting in release of free secretory component. The secretory component has protective effects of its own, potentially blocking epithelial adhesion of enterotoxigenic E. coli and neutralizing the effects of other pathogens [148] . Expression of pIgR in the mammary gland is under control of hormones responsible for initiation of lactation [153] . Elevated transport of IgA also may occur during mammary gland involution in cattle and persist longer into the involution process [140] . Part of the transfer of passive immunity story in mammals involves the timing and location of transfer of immunoglobulins from the mother to the offspring, while another part encompasses the fate and function of the immunoglobulins once in the neonate [1, 7, 127] . In humans, intestinal transfer of maternal IgG from colostrum is sparse in the neonate and their immune competency is assured by transfer via the placenta. In rats and mice, there is FcRn-mediated uptake of IgG from the colostrum and milk in the neonate intestine. In ungulate species such as cattle, sheep, goats and pigs, the young are born essentially agammaglobulinemic and rely entirely on uptake of colostral immunoglobulins, especially IgG, for systemic immune protection. The consumption of colostrum by the neonatal calf has significant effects on the gastrointestinal tract [154] . The intestinal uptake in the immediate period after birth is transient and nonselective in species such as cattle, sheep, goats, swine and others. The intestinal cells become unable to absorb macromolecules within 24-36 h after birth probably as a result of developmental processes occurring in the enterocytes [155] . The process whereby the intestinal cells gradually stop absorbing macromolecules is termed -closure‖. Before closure, the enterocytes will nonselectively absorb large molecular weight proteins and other molecules [155] . Macromolecules so transported are released into the lamina propria and then are absorbed into the lymphatic or portal circulation. Failure of passive transfer of immunity in these species is defined as occurring when a threshold concentration of IgG is not reached before closure occurs, which in the calf corresponds to serum IgG levels less than 10 mg/mL [156] . The maternal IgG in the calf's blood gradually declines over the initial month after birth, and has a half-life of approximately 16 days [157] . Milk sIgA is not taken up by the infant's intestinal mucosa [148, 158] . In fact, gut closure in humans occurs before birth and little immunoglobulin is absorbed intact in the intestine after birth [148, 158] . However, the presence of sIgA in the intestinal lumen is part of the protective function of the epithelial barrier in the intestine [159] . Milk sIgA in the intestine will bind bacteria, toxins and other macromolecules, limiting their ability to bind to intestinal cells and thereby be transported through the mucosa to the lamina propria to cause a systemic immune response [160] . In adults of a pIgR-deficient strain of mice, which do not transport sIgA into the intestinal lumen, there is an increased serum IgA and IgG that react with commensal organisms and food antigens [161] . This may be occurring because sIgA is not being secreted into the intestinal lumen to participate in its role in immune exclusion (see section 6.3), and resulting in an increased uptake of food antigens and microbial antigens from the intestinal lumen which pass to the lamina propria and stimulate specific antibody responses [161] . Development of the GALT system is dependent on microbial stimulation [148, 158] . The microbe binding function of sIgA then modulates the early microbial colonization of the gastrointestinal tract and the interaction of those microbes with the developing neonatal immune system [148, 158, 160] . From the discussion of immune milk products above it was clear that these products have protective effects on neonatal health, as well as infant and adult human health. The exact mechanisms by which immune milk products have their effects are less clear and deserve further investigation. Below are summarized several perspectives to consider when evaluating the effects of immune milk products and the role of immunoglobulins in achieving those effects. It should be remembered that colostrum and milk not only contain immunoglobulins, but also contain a range of antimicrobial factors and factors that may impact the immune system [10, 154, 160, [162] [163] [164] [165] [166] [167] [168] . These include the iron-binding antimicrobial protein lactoferrin, antibacterial enzyme lactoperoxidase, antibacterial and lytic enzyme lysozyme, oligosaccharides that function as analogues of microbial ligands on mucosal surfaces, antimicrobial heat stable peptides (defensins), and soluble CD14. In addition, colostrum and milk contain leukocytes, including activated neutrophils, macrophages and lymphocytes. Colostrum also contains cytokines and growth factors that may affect neonatal intestinal development, as well as intestinal immune responses to disease in adults [166, 169] . The relative concentrations of these factors vary considerably among species. Furthermore, colostrum provides a source of energy which may impact IgG absorption in the neonate [170] , and provide additional energy for an effective immune response. Another point to consider is that, while most macromolecules are degraded by digestive enzymes, some portion of macromolecules is transported across the intestine intact, including proteins [171, 172] . Much of the immunoglobulin consumed in an immune milk can be expected to be partially or completely digested (discussed in section 7.3), however some portion of the immunoglobulin will remain intact or at least partially intact and capable of binding to an antigen. All colostrum and milk will contain some sIgA, even those collected from cattle. The sIgA present in these secretions may contribute to the protective effects of immune milk products. Secretory IgA is considered to be the primary immunoglobulin responsible for immune protection of mucosal surfaces such as the intestine [158] . Secretory IgA and sIgM, as polymeric forms of the respective immunoglobulins, are stabilized by their binding to SC. They have antimicrobial properties such as agglutination of microbes and neutralization of viruses, and noninflammatory extracellular and intracellular immune exclusion by inhibiting adherence and invasion of mucosal epithelia [158] . The intracellular immune exclusion occurs when sIgA is being transcytosed by the enterocytes and comes into contact with viral particles within the endosomic system [15] . Secretory IgA also neutralizes pathogens in the intestinal lumen [173] . Bacterial enterotoxins may be neutralized by binding sIgA and internalization into intestinal epithelial cells [174] . In addition, IgA has a major role in the immunosuppressive mechanisms in the intestine that inhibit proinflammatory responses to oral antigens, which is part of the oral tolerance mechanisms in the intestine [158] . This suppression of the proinflammatory mechanisms is counterbalanced by systemic immune factors, including systemic IgG, which may result in inflammation and tissue damage once an antigen crosses epithelia barrier to the lamina propria [158] . After closure, any IgG localized in the lamina propria, whether from systemic sources or from uptake from the intestinal lumen, could contribute to proinflammatory responses in the intestine [158] . Indeed, post-closure uptake of IgG can occur via the FcRn receptor. FcRn has been identified in the human adult intestine [175, 176] , consistent with the hypothesis that FcRn is involved in IgG recycling (discussed in section 5.1). However, the transport of IgG across the enterocyte seems to be bidirectional, lending support to the concept that IgG in the intestine is involved in immune surveillance and defense of the mucosal lining [176] [177] [178] . Intestinal FcRn may deliver IgG-antigen immune complexes to the lamina propria for immune processing [158, 177] , thereby enhancing local mucosal immune response. On the other hand, functionally intact IgG that remains in the intestinal lumen might be expected to bind antigens and participate in protection of the tissue through immune exclusion. The intestinal mucus layer does provide an important protective barrier in the interactions of the intestinal tissue with microbes [179] . Interestingly, an IgG Fc binding site has been identified in association with the intestinal mucus [180] [181] [182] . This IgG Fc binding protein is distinct from the FcRn receptor. The Fc binding protein may block passage of IgG-antigen complexes to the enterocyte surface, thereby blocking their uptake and transport to the lamina propria, and perhaps allowing the complexes to be degraded in the intestinal lumen and excreted [169, 182] . Consumed colostrum also may impact immunological development of the neonate [1] . These maternal antibodies may then inhibit infant responses to vaccine administration and impact development of the infant's immunity [183] . In the case of dairy animals producing colostrum or milk immunoglobulins for human consumption, immunoglobulins are harvested at milking and undergo various types of processing whether it is to prolong the shelf-life of the milk, to concentrate or isolate the immunoglobulins from the mammary secretion, or to digest the milk in the intestine. Through such processing, immunoglobulins are exposed to a number of conditions that may alter the structure and function of the protein. Some of methods used to concentrate or isolate the immunoglobulins include steps that involve exposing the protein to heat, acid or pressure which may affect the conformation of the protein, and ultimately the immunological activity of the antibody. A range of methods have been used for isolation of immunoglobulins from colostrum or milk. These include traditional methods of ammonium sulfate precipitation and column chromatography [3, 5, 145, 184] . Affinity chromatographic methods used to isolate IgG include lectins [185] ; protein A or G chromatography [186, 187] , and more recently, isolation with protein A/G immobilized electrospun polyethersulfone membranes [188] ; metal chelate chromatography [189, 190] ; and adsorption with polyanhydride microparticles [191] . The range of detection and quantification methods for IgG, most often analyzed by radial-immunodiffusion [192] or enzyme-linked immunosorbant methods [193] , are now expanding to include methods that detect multiple proteins, such as thermally addressed immunosorbant assays [194] , and rapid methods that may be integrated into milking systems, such as surface plasmon resonance-based immunosensors [195] . Pepsin is a major proteolytic enzyme produced by the stomach. Pepsin digestion of IgG yields an F(ab') 2 fragment that includes the two antigen-binding (Fab) sites of the IgG molecule [5, 114, 196, 197] . Intact immunoglobulin, F(ab') 2 and other antibody formats are being exploited in development of antibody therapeutics [198] . In the small intestine, immunoglobulins are further digested by pancreatic enzymes. One of them, trypsin, preferentially digests bovine IgG1 over IgM, whereas another enzyme, chymotrypsin, preferentially hydrolyzes IgM over IgG [199] . Bovine IgG1 is more susceptible to hydrolysis by pepsin than IgG2, while IgG2 is more susceptible to trypsin [200] . Immunoglobulins are relatively more resistant to gastrointestinal digestion than other milk or colostral proteins. Upon ingestion and entry into the stomach, the caseins form a curd under the influence of the acidic environment and proteolytic activity. As a consequence, casein is retained in the stomach of the neonate longer than the whey proteins, including IgG [201] . In the intestine, the fate for the other major whey proteins is rapid digestion for α-lactalbumin, while β-lactoglobulin is more slowly digested. Intestinal digestion of IgG is among the slowest of the whey proteins and IgG provides the smallest proportion of amino acids to the neonate relative to the other major whey proteins [201] . In vitro incubations of IgA and IgG with small intestinal content of young lambs have shown that IgA is more resistant towards digestion than is IgG [17] . In adult humans consuming a bovine whey protein concentrate, approximately 59% of IgG and IgM was detected by radial immunodiffusion from effluents from the jejunum, while 19% was detected in the ileum [202] . These estimates of digestion of immunoglobulin compare with estimates of digestion of milk proteins in adult humans which are approximately 42% complete at the end of the jejunum and 93% complete by the end of the ileum [203] , again underscoring the relative resistance of immunoglobulins to digestion in the gastrointestinal tract. Detectable immunoglobulin in stool samples of infants fed the same immune product accounted for 10% of the ingested immunoglobulin [58] . In adults fed a bovine immunoglobulin concentrate, fecal IgG was typically less than 4% of ingested dose [204] . Detectable IgG in the stool [204] , or ileal effluent samples of adults [205] , is not significantly increased by prior treatment with a proton pump inhibitor to reduce stomach acid production. However, encapsulation of the immunoglobulin product can significantly increase the IgG detectable in the stool [204] , although only low levels of IgG are detectable in the ileum of adults ingesting encapsulated immunoglobulin [205] . These studies suggest that degradation of immunoglobulins is occurring throughout the intestinal tract [202] . The primary structure of the immunoglobulin found in intestinal effluents most likely is the immunoglobulin Fab or F(ab') 2 fragments found in their stool [58, 202] , which nevertheless maintains its antigen-binding activity, as indicated by the correlation between appearance of the immunoglobulin and the virus-neutralizing activity observed in stool samples [58] . In adults ingesting bovine anti-Clostridium difficile immunoglobulins, toxin-neutralizing activity paralleled the bovine IgG content in ileal effluent [205] , and in stool samples [204] . A pepsin-resistant form of bovine IgG representing approximately 10% of colostral immunoglobulin has been isolated with a lectin that binds O-linked oligosaccharides [185] , indicating that some proportion of IgG in the gastrointestinal tract may remain intact. The pH of bovine mammary secretions transiently drops at calving (to approximately pH 6.4), then increases over several days to pH 6.6 to 6.9 [206] , which is the pH characteristic of mature milk. Therefore, bovine colostrum is slightly more acidic than mature milk. Studies of isolated immunoglobulin stability over a pH range indicate that bovine IgG isolated from milk is stable for several hours at 37 °C when in pH 6-7, however stability is significantly reduced at pH ≤ 3 and ≥10 [207, 208] . The negative effect of pH on IgG stability, even in the range of 4.5-6.5, is enhanced under elevated temperature conditions [209, 210] . The use of a multiple emulsion to encapsulate milk IgG may increase stability of the protein against extreme acidic or alkali conditions, as well as against proteolytic degradation [211] . However, emulsification by homogenization may reduce the residual IgG content of the emulsion product [211] , probably as a result of high shear forces [208] . Ultrasonic treatment of isolated IgG also decreases residual IgG content [208] . Immunoglobulins are thermolabile. Exposure to temperatures of 75 °C can reduce detectable isolated bovine IgG by 40% in 5 min, and by 100% at 95 °C for 15 s [208] . Heat exposure causes conformational changes in the IgG molecule [212] . Antigen-binding activity of bovine IgG also is reduced after heat treatment [209, 213] . This is consistent with studies that suggest that the antigen-binding region of the immunoglobulin molecule is more thermolabile than the other regions of the molecule [209, 214] . Detectable IgG in colostrum or colostral whey also are reduced by heat treatment, however at a slower rate than for isolated IgG. Thermal protectants such as sugars or glycerol can increase the stability of isolated IgG to heat treatment [208, 215] . Many milk processing protocols include heat treatment of the colostrum, milk or whey. Of the major immunoglobulin classes in bovine milk, IgG is the most thermostable and IgM is the least thermostable [214] . Commercial milk samples that have undergone a typical pasteurization process, including skim milk powder, can retain 25-75% of the IgG concentration compared with raw milk, while milk undergoing ultra-high temperature (UHT) pasteurization contains little detectable IgG [192, 216] . Nevertheless, antigen-specific IgG in milk is relatively stable under typical conditions of pasteurization when compared with that in UHT milk or cow milk-based infant formulas that undergo high-temperature processing [54, 217] . Flash-heat treatment of human breast milk, a method recommended by WHO to reduce vertical transmission of HIV in resource-poor regions, has minimal effects on milk IgA and antimicrobial activity of the milk [218, 219] . This method involves placing a jar of milk into a water bath, the water bath is heated to boiling, and then the jar of milk is removed and allowed to cool. The milk reaches a maximum temperature of 72-73 °C and is above 56 °C for over 6 min [218, 219] . Alternative methods of achieving microbial inactivation may offer a means of avoiding the impact of heat treatment on IgG solutions. For example, high voltage pulsed electric fields have been used as a nonthermal processing method for pasteurization in various foods [220] [221] [222] . Pulsed electric field processing also generates heat, however temperature exposure of the fluid is less than 50 °C, and total treatment time exposure is in milliseconds [223] . That compares with more typical pasteurization process at about 72 °C for 2 min. Microbial inactivation in bovine IgG solutions as a result of pulsed electric fields did not change the secondary structure or the thermal stability of the secondary structure of the IgG [224] , and antigen-binding activity was unchanged [223] . Another emerging technology that may provide a nonthermal microbial inactivation treatment for milk uses exposure to pulsed ultraviolet light [225] . High-pressure processing is another non-thermal method with the potential for inactivation of microbial and certain enzymes in food products, thereby extending shelf-life of the product [226] . While the high-pressure process also generates heat during the treatment of the sample, lowering the initial temperature of the sample allows for control of the maximum temperature reached to be maintained within a desired range [227] . To be effective in inactivating bacterial spores, high-pressure processing needs to be combined with moderate temperature treatment [228] . Moderate to extensive loss of immunoactivity of IgG may occur depending on the conditions used for high-pressure processing of colostrum or other IgG-containing fluids [227, 229] . High-pressure processing also has been used for human breast milk with minimal effect on the milk IgA [230] . The issue of heating effects on immunoglobulin and colostrum also is important for control of various diseases that occur in cattle. Collection and storage of colostrum from dairy cows shortly after calving has long been a common procedure. The stored colostrum then is fed to newborn calves to assure adequate uptake of IgG for protection of the calf. Several pathogens can be transmitted from cow to calf via colostrum or milk [231] . Colostrum may contain these pathogens as a result of shedding from the mammary gland, contamination of the colostrum after harvesting or improper storage of colostrum prior to feeding calves [231] . One approach to allowing the neonate the benefits of colostrum from infected cows is collecting colostrum and batch pasteurization of pooled colostrum prior to feeding to the calves [232] . Volume of the batch of pooled colostrum that is pasteurized affects measurable IgG concentrations in the colostrum, as well as IgG serum concentrations attained in calves after feeding the colostrum [232] . Heat treatment of colostrum at 60 °C for one to two hours does not alter measurable IgG concentrations or viscosity of the colostrum, nor does that treatment affect antibody activity [231, 233] . In addition, bacteria inoculated into colostrum prior to a heat treatment of 60 °C for one hour are not detectable after the heat treatment [234] . On-farm heat treatment of colostrum (60 °C for one hour) results in higher concentrations of serum IgG and greater apparent absorption efficiency of IgG in new born calves consuming the treated colostrum than consumption of raw colostrum [235] [236] [237] . Colostrum and milk are rich sources of immunoglobulins. These secretions have developed through evolution to ensure homologous transfer of passive immunity from mother to offspring. The immunoglobulins that are passed from mother to her offspring, whether by transplacental transfer or by ingestion of colostrum and milk, can form an important link between the immunological experience of the mother and the immune capacity of the newborn. This immunological link also includes many immune factors that may be present in mammary secretions other than the immunoglobulins. The immunoglobulins in colostrum and milk also provide opportunities to harness their immunological function for the benefit of other animals, including humans. Research has demonstrated that bovine colostrum and milk, whether or not they are from cows immunized against specific pathogens, provide a medium for the heterologous transfer of passive immunity, and may offer disease protection in a range of species. New technologies for enhancing efficacy of vaccination, enhancing stability and extending shelf-life of the immunoglobulin preparation while minimizing the impact of the processing, and extending the effectiveness of the immunoglobulin in the intestine, may enhance future use of colostrum and milk based on their potent immunological activity. While the mechanisms by which immunoglobulins are transferred from mother to neonate and their role in the neonate have become well documented, additional research is needed to clarify the mechanisms of action of the immunoglobulins derived from milk or colostrum when used in animals that are developmentally more mature.

The crazy-paving pattern: a radiological-pathological correlation

The crazy-paving pattern is a linear pattern superimposed on a background of ground-glass opacity, resembling irregularly shaped paving stones. The crazy-paving pattern is initially described as the pathognomonic sign of alveolar proteinosis. Nowadays this pattern is a common finding on high-resolution CT imaging, and can be seen in a number of acute and chronic diseases. The purpose of this paper is to illustrate different diseases that cause this crazy-paving pattern and to correlate the radiological findings from computed tomography with the histopathological findings.

The superimposition of a linear pattern on ground-glass opacity on computed tomography images results in a pattern that is termed crazy-paving pattern, resembling the structure of irregularly shaped paving stones [1, 2] . The crazy-paving pattern is a common finding on thin-section computed tomography (HRCT), but also on multidetector computed tomography (MDCT). Ground-glass opacity is defined as a hazy increase in lung density with preservation of airway and vessel margins [3] . Ground-glass opacity occurs when there is a mild decrease in the amount of air in the airspaces and a filling of the airspaces with fluid, cells or other material, thickening of the alveolar walls or thickening of the interstitium. The linear component of this pattern can be caused by a thickening of the interlobular septa (septal lines), a thickening of the intralobular septa and the intralobular interstitium (intralobular reticular pattern and intralobular branching lines) or a linear deposition of material within the airspaces at the borders of the acini (periacinar pattern) ( Fig. 1 ) [4] . The crazypaving pattern was initially described as a pathognomonic sign of alveolar proteinosis; however, nowadays, this pattern has been reported in a variety of acute and chronic diseases as summarised in Table 1 [2, [5] [6] [7] [8] . The purpose of this paper is to illustrate different diseases showing a crazypaving pattern. The diagnosis is made based on clinical or on histological findings. If histopathological proof is available, a radiological-histopathological correlation is made. A retrospective review of the medical records of our radiological computed tomography database was performed, from 1 January 2008 until 31 December 2008, searching for patients reported to have a "crazy-paving" pattern on a CT of the chest. In total, 98 patients with a crazy-paving pattern were retained and reviewed. To rule out acute pulmonary embolism, most of the patients underwent interstitial pathological features or underwent their chest CT in an oncological setting. All these patients underwent a dedicated MDCT of the chest with 100 mAs, Fig. 1 a Anatomy of the secondary pulmonary lobule. b-e The reticular pattern: b thickening of the interlobular septa; c thickening of the intralobular interstitium; d irregular areas of fibrosis; e periacinar pattern 120 kV, slice thickness of 1 and 3 or 5 mm, and table feed of 12 mm per rotation, with or without intravenous contrast administration, according to the indication of chest CT. Only seven patients with a crazy-paving pattern on chest CT also underwent an open lung biopsy to make the definitive diagnosis. In 59 patients, the definitive diagnosis was made on a clinical basis. In the remaining 32 patients, the cause of the crazy-paving pattern remained undecided, because patients were not followed further in our institution. Ninety-eight patients with a crazy-paving pattern were retained and reviewed. Table 2 summarises the different causes of the crazy-paving pattern as found on open lung biopsy or based on clinical decision. Only seven patients underwent open lung biopsy to establish the diagnosis. A 46-year-old man presented with a 1-week history of progressive dyspnoea. He also complained of a cough and the production of white mucus in the morning. He reported a smoking habit of one pack of cigarettes per day with no further information regarding his past smoking history. Chest radiograph and CT were undertaken. Chest radiograph (Fig. 2a) showed a reticular pattern more pronounced in the central parts of the lungs. There was also an increase in lung density centrally in both lungs. No pleural fluid was noted, and the heart and central vascular structures were normal. On CT, there was a patchy distribution of areas with increased lung attenuation throughout both lungs. Superimposed on this increased lung attenuation a linear pattern was seen. There were multiple small regular and irregular lines. Some of them were thickened interlobular septa. More lines were visible in the centre of the secondary pulmonary lobule in a very irregular pattern suggesting thickening of the intralobular interstitium. Histopathological evaluation of a specimen from open lung biopsy out of the right lung showed amorphous eosinophilic material in the alveoli, positive on periodic acid Schiff (PAS) staining. This eosinophilic material corresponded with deficient surfactant (Fig. 2c) . The lines visible on CT corresponded to deposition of material within the airspaces at the borders of the acini in the secondary pulmonary lobules (periacinar pattern; Fig. 2b ). The diagnosis of alveolar proteinosis was made. A 62-year-old woman with progressive shortness of breath on exercise. Chest radiograph and CT were undertaken. Chest radiograph showed a patchy distribution of areas with increased lung density (Fig. 3a) . There was also an increase in linear markings in both lungs. On CT, a crazy-paving pattern was seen with a geographic distribution. Some of the lines were thickened interlobular septa. Centrally in the secondary pulmonary lobule we could also see a spider of lines: thickening of the intralobular septa. These findings were seen predominantly in the upper lung areas (Fig. 3b) . Although the patient had no history of bird exposure, serum precipitins against pigeons were elevated. To resolve this paradox, an open long biopsy was performed. Histology demonstrated interstitial pneumonia with lymphocytes, plasma cells and foamy macrophages in the interstitium. Epithelioid granulomas without caseation were also seen. There was no fibrosis (Fig. 3c) . The diagnosis of hypersensitivity pneumonitis was made. An 80-year-old man with rapidly progressive dyspnoea. A chest radiograph and CT were undertaken. The chest radiograph showed a patchy distribution of areas with consolidation. There was also a fine reticular pattern, most pronounced in the periphery of both lungs (Fig. 4a ). Chest CT, performed to rule out acute pulmonary embolism, was negative for the presence of lung emboli. A crazy-paving pattern with a scattered distribution of ground-glass opacities and a linear pattern superimposed, with multiple small irregular lines, was visible. Traction bronchiectasis was seen in the periphery of both lungs (Fig. 4b) . Chest radiograph showed a reticular pattern that was most pronounced in the central parts of the lungs. There was also a decrease in the lung translucency centrally in both lungs. Heart and central vessels were normal. There was no pleural effusion. b On CT, a patchy distribution of a crazy-paving pattern was visible. The lines corresponded to a deposition of material within the airspaces at the borders of the acini (1) in the secondary pulmonary lobules, but also along the interlobular (2) and intralobular septa (3): the periacinar pattern. c Radiological-histopathological correlation. Histopathological evaluation of a specimen out of the right lung showed amorphous eosinophilic material in the alveoli (*) positive on periodic acid Schiff (PAS) staining. This material corresponded to deficient surfactant. Filling of the alveoli (*) was responsible for the ground-glass appearance on CT. When the airspaces adjacent to the inter-and intralobular septa (black arrow) and to the alveolar walls filled, the periacinar pattern became visible On histology, thickening of the interstitium with variable degrees of severity was seen, leaving some alveolar septa almost completely normal, whereas others were thickened. Fibrinous exudates, honeycombing and mild inflammatory alveolitis were also present (Fig. 4c) . The diagnosis of usual interstitial pneumonia (UIP) was made. A 56-year-old woman with increasing dyspnoea. A chest radiograph and CT were undertaken. The chest radiograph showed a reticulation of the lung parenchyma, diffusely spread in both lungs, centrally and peripherally (Fig. 5a) . Chest CT showed a crazy-paving pattern especially in the periphery of both lungs. There was an increase in lung Fig. 3 Hypersensitivity pneumonitis. a Chest radiograph showed patchy distribution of areas with increased lung density. There was also an increase in the linear pattern in both lungs. b On CT, a crazy-paving pattern was seen with a geographic distribution of ground-glass opacities with the superimposition of thickened inter-(1) and intralobular (2) septa. The findings were seen predominantly in the upper lung areas. c Radiological-histopathological correlation. Histology demonstrated interstitial pneumonia with lymphocytes, plasma cells and foamy macrophages in the interstitium. Epithelioid granulomas without caseation were also seen. There was no fibrosis. The alterations in the walls of the alveoli and the inflammation in the interstitium were visible as thickening of the inter-and intralobular lines Fig. 4 Usual interstitial pneumonia. a Chest radiograph showed patchy distribution of areas with consolidation and a fine reticular pattern, most pronounced in the periphery of both lungs. b A crazy-paving pattern was visible with scattered distribution. Superimposed on the ground-glass opacities a linear pattern with multiple small irregular lines was visible (intralobular fibrosis) (1). Traction bronchiectasis was seen in the periphery of both lungs (white arrow). c Radiologicalhistopathological correlation. On histology, thickening of the interstitium (arrow) with variable severity was seen, leaving some alveolar septa almost completely normal, whereas others were thickened. Fibrinous exudates, honeycombing (*) and mild inflammatory alveolitis were also present Fig. 5 Non-specific interstitial pneumonia. a Chest radiograph showed reticulation in the lung parenchyma, diffusely spread in both lungs, centrally and peripherally. b Chest CT showed a crazy-paving pattern especially at the periphery of both lungs. There was an increase in lung attenuation (ground-glass opacification) with a superimposition of a reticular pattern with thickening of the inter-(1) and intralobular (2) septa. c Radiologicalhistopathological correlation. Histological evaluation showed a homogeneous fibrotic thickening of the interstitium with inflammation. Macrophages were visible within the alveolar septa. Homogeneous interstitial inflammation was seen, corresponding to the diffuse ground-glass opacities, whereas fibrosis in the interstitium and alveolar septa (black arrow) was related to the superimposed linear pattern Fig. 6 Radiation pneumonitis. a Chest radiograph showed an area of consolidation in the right lung with an air bronchogram. There was also loss of volume of the right lung. b CT showed the therapy response of the tumour. There was patchy distribution of a crazy-paving pattern with increased lung attenuation (ground-glass opacity) and thickening of the interlobular septa in the right lung (1) . c Radiologicalhistopathological correlation. Histological examination after autopsy showed airspace filling with an exudate in combination with thickening of the interlobular septa (arrow), thickening of the interstitium surrounding the airspaces and also the presence of irregular fibrosis (dotted arrow). Alveolar spaces filled with an exudate of proteinaceous material were responsible for the ground-glass opacities on CT. The reticular pattern was due to congestion of capillaries and oedema of the interstitium Fig. 7 Exogenous lipid pneumonia. a Chest radiograph showed a decrease in lung translucency in the caudal region of the right lung with an air bronchogram. b Chest CT showed a crazy-paving pattern with areas of increased lung attenuation and with thickening of interlobular septa (1), even thickening of the intralobular interstitium (2) . c Radiologicalhistopathological correlation. Histological examination showed alveoli filled with lipid particles (*), some ingested in macrophages (+) with the formation of lipid granulomas Fig. 8 Lymphangitic carcinomatosis. a Chest radiograph showed a pleural effusion in the right haemothorax. An increased linear pattern was seen in the left and right upper lung. b CT showed a diffuse crazy-paving pattern with areas of groundglass attenuation and thickening of the interlobular septa (1). There were also some small nodular lesions visible mostly in the left upper lobe suggestive of pulmonary metastases (2) . c Radiological-histopathological correlation. Histological examination of the autopsy specimen demonstrated thickening of the interlobular septa (*) due to fibrosis and the presence of tumour cells. There was also perivascular (arrow) thickening due to an expansion of lymphatic spaces by tumour cells. The histological reaction was that of diffuse alveolar damage and consisted of hyaline membranes in the alveolar ducts and respiratory bronchioles while the alveolar spaces fill with an exudate of proteinaceous material. This corresponded to the ground-glass opacities on CT. The reticular pattern was due to congestion of capillaries and oedema of the interstitium attenuation (ground-glass opacification) with a superimposition of thickened inter-and intralobular septa (Fig. 5b) . Histological evaluation showed a homogeneous fibrotic thickening of the interstitium with inflammation. Macrophages were visible within the alveolar septa (Fig. 5c) . The diagnosis of non-specific interstitial pneumonia (NSIP) was made. A 71-year-old man with a limited small cell lung cancer developed fever and a cough after radiation therapy. A chest radiograph and CT were undertaken. The chest radiograph showed an area of consolidation in the right lung with an air bronchogram. There was also loss of volume of the right lung (Fig. 6a) . CT showed a decrease in the size of the tumour consistent with response to therapy. There was a patchy distribution of a crazy-paving pattern with ground-glass opacities and thickening of the interlobular and intralobular septa (Fig. 6b) . Histological examination after autopsy showed airspace filling with an exudate in combination with thickening of the interlobular septa, thickening of the interstitium sur-rounding the airspaces and also the presence of irregular fibrosis (Fig. 6c) . The diagnosis of radiation pneumonitis was made. A 54-year-old man with progressive dyspnoea. A chest radiograph and CT were undertaken. The chest radiograph showed decreased translucency with an air bronchogram in the right lower lobe. There were no signs of interstitial lung disease (Fig 7a) . Chest CT showed a crazypaving pattern with areas of increased lung attenuation and with thickening of the interlobular septa, even thickening of the intralobular interstitium in the right middle and lower lobe (Fig. 7b) . Histological examination showed alveoli filled with lipid particles, some of them ingested in macrophages with the formation of lipid granulomas (Fig.7c) . The diagnosis of exogenous lipid pneumonia was made. A 73-year-old woman with an insidious onset of unexplained and progressive dyspnoea. Acute respiratory distress syndrome. CT revealed bilateral areas with ground-glass attenuation superimposed with a reticular pattern. These lines corresponded to thickening of the interlobular septa, but also thickening of the intralobular interstitium A chest radiograph and CT were undertaken. The chest radiograph showed a pleural effusion in the right hemothorax. An increased reticular pattern was seen in the left upper lung field and to a lesser degree also in the right upper lung field (Fig. 8a ). CT showed a diffuse crazypaving pattern with areas of ground-glass attenuation and thickening of the interlobular septa. There were also some small nodular lesions visible, mostly in the left upper lobe, suggestive of pulmonary metastases (Fig. 8b) . Histological examination of the autopsy specimen demonstrated heterogeneous thickening of the interlobular septa due to fibrosis and the presence of tumour cells. There was also perivascular thickening due to an expansion of lymphatic spaces by tumour cells. The diagnosis of lymphangitic carcinomatosis was made. A 34-year-old woman with thrombotic thrombocytopaenic purpura and severe myasthaenia gravis developed progressive respiratory insufficiency. A chest CT was undertaken. CT showed a patchy distribution of areas with ground-glass opacification in both lungs, more pronounced in the central parts of both lungs (Fig. 9) . There was also a superimpo-sition of a linear pattern. Most of the lines were thickened interlobular septa. The diagnosis of pneumocystis jirovecii pneumonia was made based on clinical and laboratory findings. A 67-year-old woman who received a total knee prosthesis developed septic shock with ARDS in the postoperative period. A chest CT was undertaken (Fig. 10) . This CT revealed bilateral areas with ground-glass attenuation superimposed with thickened interlobular septa but also thickening of the intralobular interstitium. An 83-year-old man with the diagnosis of acute lymphatic leukaemia developed cardiac decompensation with oedema of the lower limbs. CT of the chest (Fig. 11) showed a patchy distribution of areas with ground-glass opacification. A superimposed linear pattern was also present. Most of the lines were thickened interlobular septa. Within the secondary pulmonary lobule, enlarged vascular structures with a spider configuration were seen. There were also some other intralobular lines. Fig. 11 Pulmonary oedema. CT showed patchy distribution of areas with ground-glass opacification and a linear pattern. Most of the lines were thickened interlobular septa. Within the secondary pulmonary lobule, enlarged vascular structures with a spider configuration were seen. There were also some other intralobular lines Fig. 12 Sarcoidosis. CT showed a diffuse increase in lung attenuation (ground-glass attenuation) with the superimposition of an irregular reticular pattern: thickening of the interstitium and thickening of the peribronchovascular interstitium A 44-year-old man with sarcoidosis underwent a control CT of the chest. There was diffuse increased lung attenuation with the superimposition of multiple irregular lines and also irregular thickening of the bronchovascular bundles: the crazy-paving pattern (Fig. 12) . Interstitial fibrosis was the cause of the irregular thickening of the interstitium. A 40-year-old man with haematopoietic stem cell transplantation. He developed dyspnoea, and CT was undertaken. CT revealed multiple areas of ground-glass attenuation and consolidations. There was also a superimposition of multiple lines: thickened inter-and intralobular septa in intralobular lines caused by fibrosis (Fig. 13) . The diagnosis of graft-versus-host disease was made. A 24-year-old woman with bilateral lung transplantation. A control CT was performed and showed patchy distribution of areas of ground-glass opacification with the superimpo-sition of thickened interlobular septa: the crazy-paving pattern (Fig. 14) . The diagnosis of organising pneumonia was made on a clinical basis. A 75-year-old man known to have bronchioloalveolar carcinoma. Chest CT showed a patchy distribution of areas with increased density, areas of ground-glass opacification and areas with consolidation. Superimposed on these areas there was a reticular pattern (Fig. 15 ). These lines correspond to a thickening of the interstitium. The diagnosis was made based on biopsy, which was not performed in our institution. The crazy-paving pattern is a non-specific pattern. Initially, this pattern was considered to be highly suggestive of alveolar proteinosis. Nowadays, we can find this pattern in different lung diseases: airspace diseases and interstitial diseases [8] . The crazy-paving pattern consists of scattered or diffuse ground-glass attenuation with superimposition of Fig. 13 Graft-versus-host disease. CT revealed multiple areas of ground-glass attenuation and consolidations. There was also a superimposition of multiple lines: thickened inter-and intralobular septa and intralobular fibrosis Fig. 14 Organising pneumonia. CT showed patchy distribution of areas of ground-glass opacification with the superimposition of thickened interlobular septa a linear pattern. These lines can be: thickened interlobular septa (septal lines), thickened intralobular septa and thickening of the intralobular interstitium (intralobular reticular pattern and intralobular branching lines), or it can be a linear deposition of material within the airspaces at the borders of the acini and the secondary pulmonary lobules (periacinar pattern) [4] . Alveolar proteinosis and exogenous lipid pneumonia are airspace diseases. In alveolar proteinosis, airspaces are filled with a phospholipoproteinaceous material. On CT, the filling of the alveoli is responsible for the ground-glass appearance. When the airspaces adjacent to the inter-and intralobular septa and to the alveolar walls fill, the periacinar pattern becomes visible (Fig. 2c) [9] [10] [11] . Exogenous lipid pneumonia is the result of chronic inhalation of oily substances and is primarily a disease that affects the alveolar spaces. On CT, diffuse ground-glass opacities and consolidations, sometimes with fat attenuation caused by large lipid particles and numerous lipid-laden macrophages distending the alveolar spaces, can be seen, especially in the lower lung areas (Fig. 7c) [9, 12, 13] . Pneumocystis jirovecii pneumonia is a common pulmonary infection in severely immunocompromised patients. Our patient was receiving treatment with Neoral, Imuran, Medrol and Mestinon. Chest radiographs can be normal in up to 18% of patients. Typical radiographic manifestations on CT are bilateral, perihilar reticular and poorly defined ground-glass opacities with superimposition of lines, which can be associated with interlobular septal thickening [14] . As described by Rossi et al. histological features contributing to the ground-glass attenuation include the foamy nature of the alveolar exudates and thickening of the alveolar walls by oedema and cellular infiltrates [2] . Hypersensitivity pneumonitis, UIP , NSIP, radiation pneumonitis and lymphangitic spread of carcinoma are interstitial diseases. In hypersensitivity pneumonitis, antigen-antibody complexes around the microvasculature cause a neutrophil-rich inflammatory response and subsequent tissue injury. Biopsy in the subacute phase shows heavy infiltrates of lymphocytes and plasma cells in the walls of the alveoli in combination with poorly formed granulomas containing foreign body giant cells. In chronic phases, the interstitial inflammation remains, but fibrosis becomes more apparent and honeycombing can occur. On CT, the alterations in the walls of the alveoli and the inflammation in the interstitium are visible as thickening of the inter-and intralobular lines and thickening of the intralobular interstitium (Fig. 3c) [15] . The cardinal features of UIP on CT include subpleural reticular opacities (intralobular and interlobular septal lines) and honeycombing, increasing from the apex to the base. Ground-glass opacities are inconspicuous or absent in UIP, but focal areas of GGO may be present [16] . On histology, the hallmark is a geographically and temporally heterogeneous parenchymal fibrosis against a background of continuing mild inflammation (Fig. 4c) [17] . In NSIP the predominant finding on HRCT is subpleural, patchy, ground-glass opacification [18] . Traction bronchiectasis, subpleural microcystic honeycombing and irregular linear opacities can be seen in more advanced cases. On histology, homogeneous interstitial inflammation is seen, corresponding to the diffuse ground-glass opacities, whereas fibrosis in the interstitium is related to the superimposed linear pattern (Fig. 5c) [19] . The inflammation of lung tissue, secondary to radiation therapy, is localised in the tissue within the radiation field and depends on the interval since completion of treatment. In the acute phase (4 to 12 weeks after completion of radiation therapy), the histological reaction is that of diffuse alveolar damage and consists of hyaline membranes in the alveolar ducts and respiratory bronchioles while the alveolar spaces fill with an exudate of proteinaceous material. This corresponds to the ground-glass opacities typically manifesting on CT. The reticular pattern that can be seen in this phase is due to congestion of capillaries and oedema of the interstitium (Fig. 6c) [20] . Pulmonary lymphangitic carcinomatosis is a metastatic lung disease characterised by diffuse spread of tumour to the pulmonary lymphatic system. When tumoral cells spread to the pulmonary lymphatic system and peri- Fig. 15 Bronchioloalveolar carcinoma. CT showed patchy distribution of areas with ground-glass opacification and areas with consolidation. Superimposed on these areas there is a reticular pattern corresponding to the thickening of the interstitium lymphatic interstitial tissue, interstitial thickening is seen on CT. The proliferation of these cells in combination with lymphatic dilatation contributes to this interstitial thickening (Fig. 8c) [21] . Sarcoidosis is a systemic entity characterised by the development of non-caseating granulomatous inflammation [22] . The most common parenchymal findings include irregular thickening of the bronchovascular bundles and small nodules in a perilymphatic distribution. Ground-glass attenuation and crazy-paving pattern are also described in sarcoidosis [1] . The linear pattern is caused by interstitial fibrosis. Adult respiratory distress syndrome (ARDS) is a form of pulmonary oedema. Diagnosis is based on impaired diffusion capacity, reduced compliance of the lung and typical radiological findings. Chest CT features are bilateral consolidation and ground-glass attenuation [23] . Other findings such as reticular and linear opacities can also be seen. Histological features include oedema of the alveoli and perivascular spaces with filling of the alveoli by a protein-rich fluid [22, 24, 25] . The progress to architectural distortion and honeycombing with thickening of the inter-and intralobular septa is responsible for the linear accentuation. The CT findings in patients with leukaemia consist mainly of ground-glass attenuation, centrilobular nodules and thickening of the bronchovascular bundles in the peripheral lung. The combination of ground-glass opacities and the thickening of the bronchovascular bundles can produce the crazy-paving pattern [26] . Accumulation of fluid in the alveolae causes the ground-glass opacification. Accumulation of fluid along the interlobular septa and along the walls of the alveolae can cause the periacinar pattern. More than half of allogeneic haematopoietic stem cell transplant (HSCT) recipients develop graft-versus-host disease (GVHD), which remains a major cause of morbidity and mortality. HRCT findings in patients with GVHD are non-specific: diffuse interstitial and alveolar infiltrates are the most prominent features [27] . Depending the interstitial and alveolar component, a crazy-paving pattern can also be seen. On biopsy multiple hyaline membranes and fibroproliferative alterations can be seen, caused by the interstitial fibrosis and responsible for the linear pattern on CT. The multiple exudates into the alveolae are responsible for the ground-glass attenuation [28] . Organising pneumonia is a chronic inflammatory process characterised by plugs of granulation tissue in the lumen of distal small airways, often extending into the alveolar spaces, associated with an interstitial cellular response [29] . Typical CT features are scattered and asymmetric bilateral subpleural as well as peribronchovascular consolidation. A crazypaving pattern can be seen but is an uncommon finding [30] . Bronchioloalveolar carcinoma (BAC) has been classified into mucinous and non-mucinous subgroups and is characterised by a lepidic growth pattern through the airways and air spaces with preservation of the lung architecture. BAC may present with a variety of CT appearances. Features of BAC are the CT angiogram sign or air bronchograms in solitary nodules and in the periphery of larger consolidations, unifocal or multifocal ground-glass opacities, the crazy-paving pattern, and lobar or multilobar consolidation and cavitating nodules [31] . In patients with a crazy-paving pattern, the ground-glass attenuation reflects the lowdensity intra-alveolar material (glycoprotein), whereas the superimposed lines are due to infiltration of the interstitium by inflammatory or tumour cells [32] . The crazy-paving pattern on CT is a non-specific finding. It is characterised by scattered or diffuse areas of groundglass attenuation with superimposition of a linear pattern. This linear network can be caused by thickening of interlobular or intralobular septa or the presence of intralobular fibrosis, or it can be caused by a linear deposition of material within the airspaces. Most diseases can be diagnosed based on clinical and radiological findings. In a minority of cases a biopsy with histopathological examination is needed to establish the diagnosis.

Comparison of percutaneous radiofrequency thermal ablation and surgical resection for small hepatocellular carcinoma

BACKGROUND: The purpose of this investigation was to compare the outcome of percutaneous radiofrequency thermal ablation therapy (PRFA) with surgical resection (SR) in the treatment of single and small hepatocellular carcinoma (HCC). METHODS: We conducted a retrospective cohort study on 231 treatment naive patients with a single HCC ≤ 3 cm who had received either curative PRFA (162 patients) or curative SR (69 patients). All patients were regularly followed up after treatment at our department with blood and radiologic tests. RESULTS: The 1-, 3- and 5-year overall survival rates after PRFA and SR were 95.4%, 79.6% and 63.1%, respectively in the PRFA group and 100%, 81.4% and 74.6%, respectively in the SR group. The corresponding recurrence free survival rates at 1, 3 and 5 years after PRFA and SR were 82.0%, 38.3% and 18.0%, respectively in the PRFA group and 86.0%, 47.2% and 26.0%, respectively in the SR group. In terms of overall survival and recurrence free survival, there were no significant differences between these two groups. In comparison of PRFA group patients with liver cirrhosis (LC) (n = 127) and SR group patients with LC (n = 50) and in comparison of PRFA group patients without LC (n = 35) and SR group patients without LC (n = 19), there were also no significant differences between two groups in terms of overall survival and recurrence free survival. In the multivariate analysis of the risk factors contributing to overall survival, serum albumin level was the sole significant factor. In the multivariate analysis of the risk factors contributing to recurrence free survival, presence of LC was the sole significant factor. The rate of serious adverse events in the SR group was significantly higher than that in the PRFA group (P = 0.023). Hospitalization length in the SR group was significantly longer than in the PRFA group (P = 0.013). CONCLUSIONS: PRFA is as effective as SR in the treatment of single and small HCC, and is less invasive than SR. Therefore, PRFA could be a first choice for the treatment of single and small HCC.

Hepatocellular carcinoma is a major health problem worldwide, with an estimated incidence ranging between 500,000 and 1,000,000 new cases annually. It is the fifth most common cancer in the world and the third most common cause of cancer-related death [1] . The prognosis of HCC is generally poor. Surgical resection (SR) remains the best hope for a cure but is suitable for only 9 to 27% of patients [2, 3] . The presence of significant background liver cirrhosis (LC) often precludes hepatic resection in patients with HCC. Recurrence in the liver remnant is also common in patients who have undergone radical hepatic resection. Currently, local ablative therapy competes with surgical resection and liver transplantation as primary treatment for small HCC. Various locoregional therapies are used to treat patients who are not candidates for surgery because of the severity of the underlying liver disease. Percutaneous radiofrequency thermal ablation (PRFA), a recently developed local ablative technique, has attracted the greatest interest and popularity because of its efficacy and safety [4] . Previous studies have shown PRFA to give good results from the perspective of tumor control, with complete tumor ablation rates of 90 to 95%, and low local tumor progression rates of 5 to 10% [5] [6] [7] [8] . Prospective randomized trials have shown PRFA to be better than percutaneous ethanol injection (PEI) in producing a higher rate of complete ablation with fewer numbers of treatment sessions [9] . However, there is still debate with regard to whether PRFA or SR is the most suitable therapy of small HCC. In the present study, we conducted a retrospective cohort study to compare the results of PRFA and SR in the treatment of small HCC. Between January 2004 and January 2010, 231 patients with single HCC ≤ 3 cm in diameter received curative treatment using PRFA or SR in our department. Before performing PRFA or SR, a full discussion was made between physician and surgeon. After giving enough information including contents of the discussion between physician and surgeon to patients, patients themselves made decisions whether they received PRFA or SR. In patients with the tumor sites extremely difficult to perform PRFA such as the site directly under the hepatic dome or the heart or with poor visibility of the tumor under ultrasonography owing to extreme obesity or impossibility of breath hold when performing PRFA, SR was performed. And in patients whom high rates of complications were expected as when tumors at the site of hepatic hilar lesion were treated by PRFA, SR was performed. In patients whom informed consent could not be obtained upon SR for the reason such as physical burden, PRFA was performed. Even in patients with poor liver function such as Child-Pugh C, if they wished to treat HCC and there were no ascites, treatment for HCC was performed after fully explaining the risk for treatment. PRFA was administered to 162 patients and 69 patients underwent SR. Written informed consent was obtained from all patients. The ethics committee of our department approved the protocols for PRFA and SR. The present study comprised a retrospective analysis of patient records and all treatments were conducted in an openlabel manner. The primary end point was overall survival and the secondary end point was recurrence free survival. HCC was diagnosed using abdominal ultrasound and dynamic computed tomography (CT) scans (hyperattenuation during the arterial phase in all or some part of the tumor and hypoattenuation in the portal-venous phase) and/or magnetic resonance imaging (MRI), mainly based on the recommendations of the American Association for the Study of Liver Diseases [10] . Arterial and portal phase dynamic CT images were obtained at approximately 30 s and 120 s, respectively, after the injection of the contrast material. Abdominal angiography combined with CT (angio-CT) assistance was performed on all patients before PRFA and SR. This was due to the fact that Yamasaki et al. reported that this technique was useful for detecting small satellite nodules [11] . Then, we confirmed the presence of single HCC ≤ 3 cm in diameter with no vascular invasion using CT during hepatic arteriography (CTHA) and arterial-portography (CTAP). With regard to the diagnosis of liver cirrhosis, resected specimen at surgery was used in the SR group, and biopsy specimen was used in the PRFA group, respectively. We routinely used a cool-tip needle (Radionics Corp., Burlington, MA, USA) while performing PRFA. Using the intercostal or subcostal approach, a 17-gauge, 2 or 3 cm cooled-tip electrode was inserted under real-time ultrasound guidance. The initial treatment was planned with one ablation for tumors of < 2 cm in diameter, and two or more ablations with the overlapping technique for tumors of ≥ 2 cm in diameter. After insertion of the electrode into the tumor, we started ablation at 60 W for the 3-cm exposed tip and 40 W for the 2-cm exposed tip. The power was increased to 120 W at a rate of 10 W/min. The duration of a single ablation was 12 min for the 3-cm electrode and 6 min for the 2-cm electrode. After PRFA exposure, the pump was stopped and the temperature of the needle tip was measured. When the temperature reached > 60°C, additional ablation was not performed. When tumor ablation was complete, thermal ablation was performed along the needle track. All patients were carefully observed for treatment-related complications. All procedures were performed under ultrasound guidance by one of five operators who had at least 3 years of experience of performing PRFA. We used the artificial ascites technique to prevent collateral thermal injury when the anticipated PRFA zone was in contact with a critical organ, such as the hepatic flexure of the colon. We also used this technique to improve visibility when the index tumor was located in the hepatic dome area. In the present study, for all patients who had received PRFA, we confirmed that the ablative margin surrounded the entire circumference of the tumor by using dynamic 16-column multi-detector CT (MDCT) using 3-mm slice scans within 1 week after PRFA and 1 month after PRFA. All procedures were performed by one of four surgeons who had at least 10 years of experience of surgical resection. Surgical resection was carried out under general anesthesia using a right subcostal incision with a midline extension. We performed anatomic partial hepatectomy with a resection margin of at least 1 cm over the tumor, based on intraoperative ultrasonography (IOUS) guidance. IOUS was routinely performed to estimate the location, size, number and feeding vessels of the tumor, as well as to give an exact vascular map of liver anatomy. The Cavitron ultrasonic aspiration (CUSA, Valley Lab Corp, USA) was used to dissect the liver tissue. Hemostasis was achieved with dipolar electric coagulation and suturing. The Pringle maneuver was usually used in case of cirrhotic liver, with a clamp/unclamp time of 15 min/5 min policy. When liver function approached normal and adverse events had disappeared after surgical resection, we permitted patient discharge. Follow-up consisted of monthly blood tests and monitoring of tumor markers, including des-γ-carboxy prothrombin, which was measured using a chemiluminescent enzyme immunoassay (Lumipulse PIVKAII Eisai, Eisai, Tokyo, Japan). Dynamic CT scans were obtained every 3-4 months after PRFA and SR. No patients were lost to follow-up. Differences between the two groups were analyzed using the unpaired t-test for continuous variables, and the categorical variables were analyzed using the χ 2 test or continuity correction method. The overall survival curves and the recurrence-free survival curves were generated using the Kaplan-Meier method and compared using the log-rank test. The relative prognostic significance of the variables in predicting overall survival were assessed using univariate and multivariate Cox proportional hazards regression models. All variables with a P value < 0.05 evaluated using univariate analysis were subjected to multivariate analysis. Results of the multivariate analysis were presented as the hazard ratio (HR) with a corresponding 95% confidence interval (CI). All statistical tests were two-sided. All data were analyzed using SPSS software, version 9.0 (SPSS Inc., Chicago, IL, USA) for Microsoft Windows. Data are expressed as means ± standard deviation (SD). Values of P < 0.05 were considered to be statistically significant. The baseline characteristics of the two groups are shown in Table 1 . Between the two groups, there were significant differences in tumor size (P = 0.001), platelet count (P = 0.004) and PIVKAII value (P = 0.037). Fifty-four of 69 patients were treated with segmentectomy; 12/69 patients received bisegmentectomy; 3/69 patients underwent hemihepatectomy. The histological diagnoses of 69 patients were as follows: well-differentiated hepatocellular carcinoma (3/69), moderately differentiated hepatocellular carcinoma (36/69) and poorly differentiated hepatocellular carcinoma (30/69). Using dynamic CT performed within 1 month after SR, we confirmed no residual HCC in the liver remnant of all patients. The mean number of treatment sessions for the 162 PRFA treated patients was 1.80 ± 0.37. Target biopsy prior to PRFA was not performed on any of the patients because of the specific complication of tumor seeding. We confirmed that all of the PRFA treated patients achieved complete ablation (ablated zone totally enveloped the tumor without enhancement) using dynamic CT prior to patient discharge and 1 month after PRFA. In the present study, there were 3 patients with Child-Pugh C who underwent PRFA. Their Child-Pugh scores were all 10 points and PRFA was performed safely in these 3 patients. The median follow-up period was 3.1 years (0.2-7 years) in the PRFA group and 3.3 years (0.7-7 years) in the SR group, respectively. Thirty-three patients (20.4%) in the PRFA group died during the follow-up period. The causes of death were HCC recurrence (24 patients), liver failure (6 patients) and miscellaneous (3 patients). Twelve patients (17.4%) in the SR group died during the follow-up period. The causes of death were HCC recurrence (9 patients), liver failure (2 patients) and miscellaneous (1 patient). The 1-, 3-and 5-year overall survival rates after PRFA and SR were 95.4%, 79.6% and 63.1%, respectively in the PRFA group and 100%, 81.4% and 74.6%, respectively in the SR group (Figure 1) . The corresponding recurrence free survival rates at 1, 3 and 5 years after PRFA and SR were 82.0%, 38.3% and 18.0%, respectively in the PRFA group and 86.0%, 47.2% and 26.0%, respectively in the SR group (Figure 2) . In terms of overall survival (P = 0.259) and recurrence free survival (P = 0.324), there were no significant differences between these two groups. Figure 1 Cumulative overall survival rate. The 1-, 3-and 5-year overall survival rates after percutaneous radiofrequency thermal ablation (PRFA) and surgical resection (SR) were 95.4%, 79.6% and 63.1%, respectively in the PRFA group and 100%, 81.4% and 74.6%, respectively in the SR group. There was no significant difference between these two groups as determined using the log-rank test (P = 0.259). We defined local tumor progression as the presence of a hypervascular nodule adjacent to the ablated area of PRFA or the resected area of SR using dynamic CT scan. 20 patients in the PRFA group and 10 patients in the SR group had local tumor progression during the observation period. The 1-, 3-and 5-year local tumor progression rates after PRFA and SR were 2.0%, 14.3% and 28.3%, respectively in the PRFA group and 2.8%, 14.3% and 22.8%, respectively in the SR group. (Figure 3 ) In terms of local tumor progression, there was no significant difference between these two groups (P = 0.746). There were 127 patients in PRFA group patients with LC and 50 patients in SR group patients with LC, respectively. The 1-, 3-and 5-year overall survival rates after PRFA and SR were 94.2%, 75.8% and 56.4%, respectively in the PRFA group with LC and 100%, 78.0% and 67.8%, respectively in the SR group with LC. (Figure 4 ) The corresponding recurrence free survival rates at 1, 3 and 5 years after PRFA and SR were 86.0%, 35.0% and 14.8%, respectively in the PRFA group with LC and 79.5%, 39.3% and 23.8%, respectively in the SR group with LC. (Figure 5 ) In terms of overall survival (P = 0.521) and recurrence free survival (P = 0.669), there were no significant differences between these two groups. There were 35 patients in PRFA group patients without LC and 19 patients in SR group patients without LC, respectively. The 1-, 3-and 5-year overall survival rates Figure 2 Cumulative recurrence free survival rate. The 1-, 3-and 5-year recurrence free survival rates after percutaneous radiofrequency thermal ablation (PRFA) and surgical resection (SR) were 82.0%, 38.3% and 18.0%, respectively in the PRFA group and 86.0%, 47.2% and 26.0%, respectively in the SR group There was no significant differences between these two groups as determined using the log-rank test (P = 0.324). after PRFA and SR were 96.6%, 87.2% and 74.4%, respectively in the PRFA group without LC and 100%, 95.6% and 95.6%, respectively in the SR group without LC. (Figure 6 ) The corresponding recurrence free survival rates at 1, 3 and 5 years after PRFA and SR were 93.0%, 52.5% and 22.2%, respectively in the PRFA group with LC and 100%, 75.7% and 30.4%, respectively in the SR group with LC. (Figure 7) In terms of overall survival (P = 0.276) and recurrence free survival (P = 0.258), there were no significant differences between these two groups. Serious adverse events were significantly more frequent in the SR group than in the PRFA group (6/69 versus 3/ 162; P = 0.023). Serious adverse events in the SR group were as follows: bile leakage (2 patients); refractory ascites (2 patients); acute respiratory distress syndrome (ARDS) (1 patient); and massive gastrointestinal bleeding (1 patient). Serious adverse events in the PRFA group were as follows: biloma (1 patient); refractory ascites (1 patient); and intra-abdominal bleeding (1 patient). The hospitalization length was significantly longer in the SR group (18.1 ± 10.4 days) than in the PRFA group (14.7 ± 5.7 days) (P = 0.013). In addition, there was no patient who died within the same hospitalization, making the mortality rate 0% in two groups. In the univariate analysis of factors contributing to overall survival, hepatitis C virus (HCV) versus non HCV (P = 0.042), serum albumin (g/dL) (> 3.5 versus ≤ 3.5) (P = Figure 3 Cumulative local tumor progression rate. The 1-, 3-and 5-year local tumor progression rates after percutaneous radiofrequency thermal ablation (PRFA) and surgical resection (SR) were 2.0%, 14.3% and 28.3%, respectively in the PRFA group and 2.8%, 14.3% and 22.8%, respectively in the SR group. There was no significant differences between these two groups as determined using the log-rank test (P = 0.746). 0.003), and platelet count (× 10 4 /mm 3 ) (> 10 versus ≤ 10) (P = 0.045) were found to be significant factors (Table 2) . However, in the multivariate analyses involving these three factors, serum albumin (g/dL) (> 3.5 versus ≤ 3.5) was the sole significant factor contributing to overall survival. Similarly, in the univariate analysis of factors contributing to recurrence free survival, HCV versus non HCV (P = 0.022), LC versus non LC (P = 0.002) and platelet count (× 10 4 /mm 3 ) (> 10 versus ≤ 10) (P = 0.005) were found to be significant factors (Table 3) . However, in the multivariate analyses involving these three factors, the presence of LC was the sole significant factor contributing to recurrence free survival. Partial hepatectomy in patients with resectable HCC, who have normal liver function and are in good general condition is still considered the gold standard therapy with the aim of delivering curability [12] . In recent years, it has been possible to reduce perioperative mortality to less than 5% depending on the extent of resection and hepatic reserve [13] . The improved outcome is primarily as a result of advances in surgical and radiologic techniques, perioperative care and more cautious patient selection [14] . Patients not eligible for resection because of their medical condition might be candidates for local ablative therapy, such as percutaneous ethanol injection (PEI) and PRFA. Many clinical trials comparing PRFA and PEI have demonstrated the clear superiority of PRFA over PEI [9, [15] [16] [17] . However, a major limitation of PRFA is the small volume of tumor that can be treated. The rate of complete ablative necrosis decreases with the size of the tumor, particularly in the case of tumors Figure 4 Cumulative overall survival rate between percutaneous radiofrequency thermal ablation (PRFA) group patients with liver cirrhosis (LC) (n = 127) and surgical resection (SR) group patients with liver cirrhosis (LC) (n = 50). The 1-, 3-and 5-year overall survival rates after PRFA and SR were 94.2%, 75.8% and 56.4%, respectively in the PRFA group with LC and 100%, 78.0% and 67.8%, respectively in the SR group with LC. There was no significant differences between these two groups as determined using the log-rank test (P = 0.521). larger than 3 cm. There is general consensus that complete response to PRFA therapy in patients is associated with improved outcome [18] [19] [20] . Therefore, in the present study, objectives were limited to patients with HCC ≤ 3 cm in size. HCC mainly disseminates through the portal and hepatic veins. The micro-dissemination can invade the tributaries of the portal branches and shed tumor emboli in the neighboring branches of the same liver segment [21] [22] [23] [24] . However, in the present study, with regard to recurrence free survival, there was no significant difference between the two treatment groups. One possible reason for this is that a sufficient ablative margin around the tumor when PRFA is administered may suppress the invasion of the micro-dissemination. Previous studies have reported that the initial treatment contributes to the survival of HCC patients treated using PRFA [19, 25] . In PRFA therapy, obtaining sufficient ablative margin around the tumor seems to be essential. The findings of the present study indicated that the overall and recurrence free survivals were the same for patients with a single HCC ≤ 3 cm in diameter treated with either PRFA or SR. In addition, PRFA was demonstrated to have an advantage over SR in causing less serious adverse events and a shorter hospitalization length. Figure 5 Cumulative recurrence free survival rate between percutaneous radiofrequency thermal ablation (PRFA) group patients with liver cirrhosis (LC) (n = 127) and surgical resection (SR) group patients with liver cirrhosis (LC) (n = 50). The 1-, 3-and 5-year recurrence free survival rates after PRFA and SR were 86.0%, 35.0% and 14.8%, respectively in the PRFA group with LC and 79.5%, 39.3% and 23.8%, respectively in the SR group with LC. There was no significant differences between these two groups as determined using the log-rank test (P = 0.669). Chen et al conducted a randomized control trial (RCT) on 180 patients with a single HCC ≤ 5 cm to receive either PRFA or surgical resection [12] , and Lu et al carried out another RCT on 105 patients with early HCC [26] . These two RCTs presented similar findings to those of our study. Additionally, four non-randomized controlled studies also reported similar findings of ours [27] [28] [29] [30] . And our study suggests that PRFA is less invasive than SR. It seems that PRFA can be a first choice for the treatment of small HCC. On the other hand, a recent study indicated that surgical resection provided better survival and lower recurrence rates than RFA for patients with HCC that conformed to the Milan criteria for a RCT [31] . However, in comparing their results with ours, the mean age of their patient population was more than 10 years younger than ours. In the etiology of liver disease in their study, patients with chronic hepatitis B were in the majority [31] . However, in our study, patients with chronic hepatitis C were in the majority. Therefore, their study results did not reflect the actual situation in Japan where Japanese HCC patients consist of many elderly patients, and the etiology of background liver disease involves chronic hepatitis C which accounts for about 80% of Japanese HCC patients. Hence, we should interpret their study results with caution. Our study had several limitations. First, it was a retrospective cohort study. Patients who had a good hepatic reserve tended to receive surgical resection, and this could have possibly led to bias. Second, we did not assess the histopathologic diagnosis of HCC in the PRFA group. Tateishi et al reported that patients with Figure 6 Cumulative overall survival rate between percutaneous radiofrequency thermal ablation (PRFA) group patients without liver cirrhosis (LC) (n = 35) and surgical resection (SR) group patients without liver cirrhosis (LC) (n = 19). The 1-, 3-and 5-year overall survival rates after PRFA and SR were 96.6%, 87.2% and 74.4%, respectively in the PRFA group without LC and 100%, 95.6% and 95.6%, respectively in the SR group without LC. There was no significant differences between these two groups as determined using the log-rank test (P = 0.276). Cumulative recurrence free survival rate between percutaneous radiofrequency thermal ablation (PRFA) group patients without liver cirrhosis (LC) (n = 35) and surgical resection (SR) group patients without liver cirrhosis (LC) (n = 19). The 1-, 3-and 5-year recurrence free survival rates after PRFA and SR were 93.0%, 52.5% and 22.2%, respectively in the PRFA group without LC and 100%, 75.7% and 30.4%, respectively in the SR group without LC. There was no significant differences between these two groups as determined using the logrank test (P = 0.258). poorly differentiated HCC had a poorer outcome than patients with well to moderately differentiated HCC after PRFA [32] . Third, our study patients were limited to patients who have undergone curative treatment. These problems should be resolved in a future prospective study. In conclusion, we demonstrated that PRFA is as effective as SR in the treatment of single and small HCC patients who have undergone curative treatment, and that PRFA is less invasive than SR. Therefore, PRFA can be a first choice for the treatment of single and small HCC.

Making sense of perceptions of risk of diseases and vaccinations: a qualitative study combining models of health beliefs, decision-making and risk perception

BACKGROUND: Maintaining high levels of childhood vaccinations is important for public health. Success requires better understanding of parents' perceptions of diseases and consequent decisions about vaccinations, however few studies have considered this from the theoretical perspectives of risk perception and decision-making under uncertainty. The aim of this study was to examine the utility of subjective risk perception and decision-making theories to provide a better understanding of the differences between immunisers' and non-immunisers' health beliefs and behaviours. METHODS: In a qualitative study we conducted semi-structured in-depth interviews with 45 Australian parents exploring their experiences and perceptions of disease severity and susceptibility. Using scenarios about 'a new strain of flu' we explored how risk information was interpreted. RESULTS: We found that concepts of dread, unfamiliarity, and uncontrollability from the subjective perception of risk and ambiguity, optimistic control and omission bias from explanatory theories of decision-making under uncertainty were useful in understanding why immunisers, incomplete immunisers and non-immunisers interpreted severity and susceptibility to diseases and vaccine risk differently. Immunisers dreaded unfamiliar diseases whilst non-immunisers dreaded unknown, long term side effects of vaccines. Participants believed that the risks of diseases and complications from diseases are not equally spread throughout the community, therefore, when listening to reports of epidemics, it is not the number of people who are affected but the familiarity or unfamiliarity of the disease and the characteristics of those who have had the disease that prompts them to take preventive action. Almost all believed they themselves would not be at serious risk of the 'new strain of flu' but were less willing to take risks with their children's health. CONCLUSION: This study has found that health messages about the risks of disease which are communicated as though there is equality of risk in the population may be unproductive as these messages are perceived as unbelievable or irrelevant. The findings from this study have implications beyond the issue of childhood vaccinations as we grapple with communicating risks of new epidemics, and indeed may usefully contribute to the current debate especially in the UK of how these theories of risk and decision-making can be used to 'nudge' other health behaviours.

Few would argue against the success of mass vaccination programmes in reducing and, in the case of smallpox, eliminating infectious diseases. Continued success however, requires adequate coverage which in turn requires parents to be committed to vaccination as an effective method of preventing their children from contracting diseases. Such commitment may be adversely affected by an increasingly complicated immunisation schedule for an increasing number of diseases and scares of vaccine safety (e.g. MMR debate that continues in the UK). It has also been argued that the very success of mass vaccination programmes has limited parents' experience of vaccine preventable diseases and thus affected their assessment of the severity of diseases and importance of prevention [1, 2] . There is therefore, a continued need to better understand parents' perceptions of what is serious, what is risky and what is best for their children's health. Indeed with SARS, bird and swine flu, this extends to what people think about how best to protect their own health. Theories of health beliefs, decisionmaking and subjective risk perception have all been used in attempts to explain parents' decisions with respect to immunisation [3] [4] [5] , but with limited success in explaining why parents differ in their perceptions of risk. With the exception of Hawe's et al research, [6] public health and health promotion campaigns have not drawn explicitly on or tested these theories. A recent review discussed how decision-making theories might help to explain parent attitudes and behaviour with respect to MMR vaccine uptake, but provided no primary evidence [7] . We argue in this paper that our understanding of immunisation choices and how we may best influence those choices may be increased through a synthesis of health beliefs, decision-making and subjective risk perception theories. This paper describes the findings from a qualitative study examining the utility of the risk perception and decision making theories to provide a better understanding of the differences between immunisers' and non-immunisers' health beliefs and behaviours, when considering the risks of a 'new strain of flu'. Theories of health protective behaviour offer an appealing framework in which to interpret differences in compliant and non-compliant parents with respect to childhood vaccinations [8] . While several models have been developed to account for people's adoption of health protective behaviour such as the Theory of Reasoned Action [9] , the Triandis Model [10] , Multi-Attribute Utility (MAU) Theory [11] and the Subjective Expected Utility Theory [8] , the Health Belief Model is possibly the simplest and the most widely used and tested [12, 13] . The four elements of the Health Belief Model are: perceived susceptibility (likelihood of getting the disease), perceived severity (perception of how serious an outcome or consequence is from the disease), perceived benefits (efficacy of preventive action undertaken) and perceived barriers (time, effort, money, inconvenience, pain, side effects of preventive action) [12, 13] . Attempts to assess the association of these elements and childhood immunisation uptake have been inconclusive with some studies reporting expected relationships [14] [15] [16] and others contrary to what would be expected [17, 18] . These contradictory findings have led to the conclusion that the health beliefs of mothers are not important contributors to immunisation uptake or completion and are less important than socio-demographic factors. That is, incomplete immunisation (fall behind the immunisation schedule or fail to complete) is associated with being poor [17] , and being a single parent [19, 20] . Low maternal education has usually been found to be a risk factor for not completing immunisation [20] . Being anti immunisation on the other hand, is associated with high education [21] . While in some instances theories of health protective behaviours have been shown to differentiate between complete, incomplete and non-immunisers, none of them provides an explanation of how people perceive risks or how their perceptions might influence behaviour. Indeed these models assume a rational basis for these decisions: a simple weighing up of information regarding severity, susceptibility, benefits and barriers. There is not, however, a simple relationship between mortality or morbidity figures and the perception of risk [22] . People do not perceive, interpret or act on risk information in the way expected by risk experts in general [22] nor specifically when considering vaccines and diseases [4, 23] . Two domains which have addressed lay rather than expert perceptions of risk and what influences decisions encompassing risk are studies of the subjective perception of risk [24] [25] [26] and the study of decision-making under uncertainty (also referred to as the psychology of choice) [27, 28] . Both approaches focus on risk as a subjective rather than an objective concept, and both involve social and psychological aspects that impact on the individual cognitive structure of risk perception. Research into the subjective perception of risk generally involves asking study participants to rate a heterogeneous set of environmental or health risks including risks from individual activities, residential or work conditions, hazards from technologies, substances or products, and natural hazards. Respondents are asked to rate these activities or hazards in terms of perceived magnitude of risk, the acceptability of the risk and other aspects such as likelihood of death, catastrophic potential, avoidability, fear, familiarity, imposed or personal choice, time scale of impact, benefits of risk source, degree of concern, personal exposure etc. Studies including vaccination (otherwise unspecified) as a hazard have reported vaccination as low risk [24, 25] . Vaccination is generally considered to be a risk that is not 'dreaded' (controllable, not fatal, individual, low risk to future generations) but is somewhat 'unknown' (not observable, effect delayed, new risk, risk unknown to science). Slovic [25] , in a study of risk perception of prescription drugs including vaccines, found vaccines were generally considered beneficial by the sample. However, people associating negative meanings to drugs tended to judge drugs and vaccines as having higher risks and lower benefits than people who associated positive meanings to drugs. This research has consistently found that risks are perceived more negatively if exposure to the hazard is involuntary, people perceive they have little personal control over outcomes and there is uncertainty about the consequences of the outcome(s), the hazard is unfamiliar, the effects of the hazard are delayed, the hazard has catastrophic potential, the benefits are not immediately apparent and the hazard is caused by human rather than natural causes [24] . These have been summarised by two factors labelled 'Dread' (uncontrollable, feared, involuntary exposure, inequitable distribution of risk, not easily reduced, catastrophic, risk increasing, fatal consequences, risk to future generations) and the 'Unknown' (not observable, risk unknown to science, delayed effect, new risk) [29] . People make judgements about the persuasiveness and trustworthiness of experts involved in communicating risk and find risks less acceptable if they believe that the communication of the risks between experts and the community is poor [30] . Research into decision-making under uncertainty showed that the rational subjective expected utility models which presumes that a good or 'rational' decision maker will sum the utilities and choose the action with the greatest total utility, does not explain how people make decisions [27, 28] . Kahneman and Tversky's research confirmed findings from the subjective perception of risk and further contributed to an understanding of how information about risks or uncertainties of outcomes influences decisions [27, 28] . Using hypothetical scenarios, studies have shown that people consistently underestimate risks of familiar and frequent events and overestimate the occurrence of low probability, but high consequence risks. Thus, rare events are perceived as more likely to occur than they do and common events are thought to occur less often than they do. The classic problems devised by Kahneman and Tversky involved asking subjects to choose between the certainty of saving 200 out of 600 lives or taking a 1 in 3 chance of saving 600 lives. They found that people's responses to possible negative outcomes are more extreme than their responses to possible positive outcomes. People also demonstrate a tendency to believe that their own risks are less than others, particularly if they believe that their exposure to risk is in some way under their control [31] . The impact of this tendency, described as unrealistic optimism and/or the illusion of control, is to reduce the perceived need to take protective measures [32] . Of particular importance to the decision to immunise are studies describing the operation of omission bias and choices involving ambiguous situations [33] . Omission bias describes a preference for taking no action if the action might cause harm, even if there is a greater risk of harm by 'doing nothing'. Ambiguity describes the decision-maker's feeling that there is missing information relevant to outcome or choice [33] . The effect of ambiguity on choice is to reduce people's willingness to act or to postpone the action until the missing information can be obtained [33] . These studies used hypothetical scenarios, making decisions about others, and participants were generally tertiary students and sometimes, parents. As stated above the Health Belief Model by itself has been found wanting in terms of being able to explain parents' behaviour with respect to immunising their children. Conceptual pieces have been written describing how subjective risk perception and risky decision-making theories may be useful in understanding parents' behaviour and choices although these have not drawn on primary data [7] . Using a qualitative design, the aim of this study was to: explore the salience of the decision-making and risk perception findings to parents' choices to immunise their young children; examine the utility of the risk perception and decision-making theories to provide a more detailed explanation of the differences between immunisers' and nonimmunisers' perceptions of severity, susceptibility to disease and benefits of vaccines; and understand how these theories might explain perceptions of risk and reactions to a 'new strain of flu' for themselves and their children. Sampling A stratified purposeful sampling strategy [34] , was used to identify first time and experienced mothers of infants who were completely immunised (for age), incompletely immunised (behind the recommended immunisation schedule), partially immunised (parents chose or advised not to have a specific immunisation) or who had no immunisations. Initially, mothers of children between 14-16 months were approached. This age allowed a range of immunisation experiences to be discussed with participants while minimising the time since the first immunisation. To obtain sufficient numbers of nonimmunisers and partial immunisers this age range was broadened to include 3 to 30 months. It was initially proposed that interviewing approximately 8 mothers from each immunisation category would be sufficient to discern patterns of similarity and difference between these categories. If preferred by the mother, both parents could participate in the interview. Possible participants were identified by Maternal and Child Health (M&CH) nurses in five metropolitan local government areas in Melbourne, Australia. Nurses were asked to approach mothers fitting the immunisation categories and to include, to the best of their knowledge, mothers of high and low education (< Year 11) and high and low income (held a Health Care Card). Parents were identified as fitting these categories from informal information available to the nurses. Parents from non-English speaking backgrounds whose English was poor were not interviewed. The Nursing Mothers Association group for the area also advertised the study. Ethics approval was granted by the Royal Children's Hospital Ethics in Human Research Committee. Participation was voluntary, with written consent required. Participants provided informed consent to be interviewed, and for the interview to be taped. Participants were assured that they could stop the interview at any time, could choose not to answer any question if they didn't want to and that their responses would be confidential and transcripts anonymised. The interviewer did not have a dual relationship with the participants (i.e. she was neither a clinician nor provider of health services/care). Recruiting and interviewing continued until 'saturation' occurred [34] (i.e. no new information was obtained from the interviews) for complete, incomplete and non-immunisers or until no more new parents fitting the categories could be identified, as was the case for partial immunisers. Over the period of data collection, 94 families were identified as possible participants. Forty-eight interviews were arranged and 45 completed. One participant withdrew consent prior to the interview (she did not believe she had anything to say). Two participants were not at home at the time scheduled for the interview and neither returned follow-up phone calls. Of those not interviewed, 17 fitted categories for which a sufficient number of interviews had been conducted (complete immunisers); 11 were not interviewed due to language difficulties; 12 could not be contacted by the nurses and 6 refused. Interviews were undertaken in 1995-96. For six interviews both mother and father participated in the interviews (three of these were non-immunising families). All interviews were conducted in the participants' homes. Semi-structured, one-on-one interviews were used to collect information from parents about their children's health, the experience of illness in the family, their understanding and interpretation of risk and how all of these related to their decision to immunise. The method of one-on-one interviews rather than focus groups, was chosen as the aim was to explore the parents' experiences and path to choosing to immunise or not, rather than a group discussion of the pros and cons of immunisation. To understand the context of parents' decisions to immunise, the interviews covered four themes: (1) how mothers keep their children healthy; (2) experience, familiarity and concerns regarding both vaccine preventable diseases and other diseases; (3) concepts and influences on risk perception and (4) the decisions, experience and outcomes regarding immunisation. The interview began with questions about health as a nonthreatening introduction and to place the consequent discussions about disease and disease prevention in the framework or context of health. Questions about diseases and the family's experience of them were included to explore the relationship between common illnesses experienced by the family and vaccine preventable diseases. What was of interest here was which diseases were familiar, which were unfamiliar, which were to be avoided if possible and which were 'just' childhood illnesses. To aid the investigation of how parents understand and interpret risk information the following two hypothetical news items about an influenza outbreak were read to the participants. Health authorities issued a warning today about a new strain of flu expected this winter. The flu affects the airways, making breathing difficult and causing repeated bouts of coughing. Long term effects of pneumonia and brain inflammation have been reported in some cases. This strain appears to affect adults between the ages of 20-50 years. Several deaths occurred last year from the A-strain of this virus. Doctors have recommended that all adults should be vaccinated, especially those who are overworked, stressed and tired. Health authorities issued a warning today about a new strain of flu expected this winter. The flu affects the airways, making breathing difficult and causing repeated bouts of coughing. Long term effects of pneumonia and brain inflammation have been reported in some cases. Several deaths occurred last year from the A-strain of this virus. Many of those who died were children under 5 years. Doctors have recommended all young children should be vaccinated. The description of symptoms and complications was taken from a description of the complications for pertussis [35] . The doctors' recommendations were written so that the parents could consider themselves 'at risk' in the 1st instance, parents of young children often feeling overworked and tired, and in the 2nd scenario their child/children fitted the 'at risk' group. Omission bias was examined in this study by asking parents to respond to the following statement: STATEMENT 1 Some people say they won't vaccinate because they would feel worse if their child died because of the injection than if the child was not immunised and died from the disease. Participants were asked their opinion and were then read a second statement: STATEMENT 2 Some people say they would vaccinate because they would feel worse if their child got the disease and died or was brain damaged when they could have had an injection to prevent it. These statements were used rather than the more complicated scenarios developed by others (e.g. [33] ) because it was believed they captured the essential element of omission bias in circumstances with which the parent could identify. The interview concluded with discussions of the process of deciding to immunise or not and included a discussion of structural and non-structural barriers. All interviews were conducted in the participants' homes at times convenient to them by the first author. Interviews lasted between 45 to 90 minutes. Socio-demographic information including family size and type (two or one parent family), mother's age, parental occupations and education levels was collected at the end of the interview. For those children who were immunised, immunisation status was determined from the immunisation records held by the parent. All interviews were audio-taped and fully transcribed. The interview focussed on the sole or youngest child in the family. Previous experience of disease and immunisations for older children was discussed in terms of its effect on decisions for the youngest child. Interviews were thematically coded after all interviews had been collected. This analysis focussed on determining whether parents' descriptions of their experiences and beliefs were congruent or incongruent with theories of health behaviour, decision-making and risk perception. The coding was undertaken by the first author. No formal testing of the reliability of the coding was undertaken although discussions with colleagues about the analysis and the meanings and patterns derived from this were extensively undertaken. Interviews were completed with 16 mothers whose children had completed immunisations appropriate for their age, 12 whose children were incompletely immunised, seven whose children were partially immunised (chose or advised not to have at least one component), and ten whose children had no immunisations. All families with incomplete immunisations had two or more children. (See Table 1 .) The following section presents a brief summary of similarities between immunisers and non-immunisers in terms of the concepts in the Health Belief Model. This is followed by a critical interpretation of the data linking this model with the theories of subjective perception of risk and decision-making under uncertainty. Finally the differences found between complete, incomplete, partial and nonimmunisers in terms of these theories are summarised. Table 2 summarises the differences and similarities between complete, incomplete and non-immunisers in terms of the core concepts of the Health Belief Model from these interviews (see [16] for further details). Partial immunisers formed two groups; those whose child had had a severe reaction to DTP (Diphtheria, Tetanus, Pertussis vaccine) (n = 3) where the parents had been advised not to continue with vaccination and those who chose to only undertake some vaccinations or changed their mind about vaccinations after the first DTP vaccine (n = 4). The former of these expressed views similar to complete immunisers and the latter to nonimmunisers. To better understand these differences in perceptions, the interviews were analysed firstly for themes from the studies of risk perception-dread, familiarity and controllability. Dread was an important determinant of what the participants in this study perceived as high risk. What was dreaded, however, differed between the immunisers and the non-immunisers. Immunisers dreaded the outcomes of the diseases, especially those with which they were unfamiliar. This fear motivated them to take the risk of immunising. "The life threatening ones really concerned me, like ones I didn't know anything about...Polio really scared me and the thought of whooping cough... those are quite scary sort of concepts. Meningitis was frightening." (Complete immuniser, #11) "I'm not sure about whooping cough, it just has horrible connotations in my mind but I'm not quite sure why, what can happen... Yeah, it's interesting isn't it, you know, people think it's the ones that you don't know about that you're likely to dismiss but it doesn't seem to me that way." (Complete immuniser, #13) Polio, diphtheria, tetanus and meningitis were unfamiliar to these mothers but they conjured vivid images of severe outcomes. Of this group (immunisers), parents considered their children to be at greatest risk from meningitis. Even though most considered it unlikely that their children would contract these diseases, it was easy to imagine that if contracted, the worst was likely to happen. LB: "If M hadn't been immunised, how likely do you think she would get these diseases?" The risks associated with vaccination were also perceived as being rare but rather than imagining the worst in this instance, they believed that one would be unlucky to have severe reactions. Thus, on balance, the risk of not immunising was not worth taking and a 'common sense' approach was necessary. They likened vaccination to taking other safety precautions. In terms of the theories of risk, respondents were perceiving the diseases as less familiar therefore dreaded and unknown, and therefore possibly overestimating their risk, and perceiving vaccines as more familiar, and possibly underestimating their risk. In contrast, non-immunisers dreaded the unknown or uncertain outcomes of the vaccines with major fears being for invisible/undetectable/distant problems such as the vaccines causing leukaemia, SIDS, AIDS and brain damage. For these parents, vaccines were not only ineffective but they were actively dangerous to children's health. "Brain damage, affecting limbs. I've read there are long term effects which we really don't know about. There are new diseases coming up. Polio is no longer life threatening but there are cancers and AIDS, long term effects on the immune system. And this is because we are interfering, causing genetic changes." (Non-immuniser #26) "They don't work and they do harm. Putting these things into their bodies-germs and all the other products-mercury aluminium etc cannot be good. It suppresses/disrupts the child's immune system. It doesn't work and it is harmful. It's not just the risk of the side effects but the long term effects that we don't know about now. Basically so many things which we did in the past we now know better and think were barbaric. I would rather not do something to my child when we don't know what the long term effects might be. There have been studies which have related these to leukaemia and other cancers, asthma eczema and all sorts of things. I don't want to do that to my child." (Non-immuniser, #19) On the other hand, severe outcomes of the diseases were believed to be rare or only a problem for children with poor nutrition, poor sanitation, and compromised immune systems. Non-immunisers believed it was unlikely that their children would suffer serious complications if they contracted these diseases because they had healthy immune systems. "If D did get one of these diseases it wouldn't necessarily be catastrophic. Some people do get very sick or die but what we don't know, what they don't tell us is that those children were probably not well to start with. Fairly sick children are more likely to get serious long term effects. Their health before the illness is crucial to how their bodies cope with the disease." (Non-immuniser,# 19) "...a rejection of the notion that children have to be immunised against these diseases because the disease itself will automatically be worse than the immunisation and a concern that the you know the vaccination itself can have problems." (Non-immuniser,# 21) For one non-immuniser, who was not 'against' using conventional medicine' she believed her children were protected from disastrous consequences because they had easy access to modern medical intervention. "Well I just can't see the need. If your child catches measles and... that develops into anything else... we're not stuck in the middle of the country without good doctors or hospitals..." (Non-immuniser, #29) As reported previously, both pro-and anti-immunisers were concerned about vaccines overloading even healthy but immature immune systems [16] "I do sort of worry that we are vaccinating too much. I just worry about what it does to your immune system, to all our immune systems" (Complete immuniser, #14) "He's still very thin but he's past the point of where I sort of see him [as] very vulnerable...like now I'm happy to give them to him." (Incomplete immuniser, #24) "I mean I know she is a strong as a horse and I know she could have every shot under the sun and she'd be fine I just don't think it's a proper thing to do ... in only two-month olds." (Non-immuniser, #29) As would be expected from risk perception theory, diseases that were familiar to parents were not dreaded. Measles, mumps and rubella were not considered serious or life threatening by most parents irrespective of immunisation status. Most mothers were familiar with these diseases. They had had personal experience of these and remembered them as mild. "[If] she happened to get measles, well I'm not that worried...because I had it and it was fine." (Complete immuniser, # 11) The motivation to immunise against these diseases was, therefore, less than for diseases that were unfamiliar. "...If it's a disease like measles, mumps, chicken pox, things like that you can let them get through fine, then if you got into meningitis, polio, well yeah, you'd have to think again." (Complete immuniser, #31) "See, measles and German measles-I know that they brought the immunisation in because there are complications and there have been kids with complications. But see, like I remember from my generation a lot of us that was just the normal. Kids had the measles. So I am not so sure about those two whether it is important." (Incomplete immuniser, #15) Measles, mumps and rubella were perceived as diseases that...'every child's got to get' and rather than avoid these diseases it was best to 'get them out of the way' as early as possible, especially as it was believed that these diseases were more serious in adults. There was no urgency in having their children vaccinated for these diseases. The idea of control was also used by parents to explain their choice to vaccinate. One explanation for immunising given by complete immunisers was that they could not control their children's exposure to diseases and hence, it was safer to vaccinate. By doing so they could control, to some extent, the diseases that their children were at risk of contracting. "I think the world of her and I thought if [I] can prevent her getting any of these diseases I will. (Complete immuniser, #7) Incomplete immunisers believed vaccination would contain or reduce the effects of disease rather than prevent it completely. "...kids still get measles and mumps so that's the silly thing isn't it really? It's only to prevent it, it can't cure, do you know what I mean? I've heard of kids still getting measles." (Incomplete immuniser #30) "But it doesn't prevent the flu, you still get the flu but not a strong dosage." (Incomplete immuniser, #44) In contrast, non-immunisers talked about being able to control their children's environment and therefore their exposure to disease. This non-immunising mother spoke of her reasons for vaccinating the family dogs: The explanatory power of ambiguity, outrage, omission bias and optimistic control was also examined in these interviews. Perceived lack of information or insufficiency of information should either provoke outrage [30] or hesitation from acting [33] . Participants provided examples of both. Lack of information about susceptibility to vaccine side effects caused mothers in some instances to refrain from vaccinating their child or to hesitate about immunisation. One mother had hesitated to immunise her second child until she could be reassured that he would not have severe side effects from the vaccine. She had reason to believe he would be particularly susceptible to such side effects because he was 'not robust', he had many food allergies and his father had collapsed after immunisation as an infant. She expressed an equivalent concern about the child's susceptibility to disease especially as he had a school-aged sibling who could expose him to disease. The major reason for her hesitation was that no one had seriously considered her questions or considered her son's case on an individual basis. "I want someone to look at him as an individual and I don't feel that they are the medical community....I don't want people making the decisions for me. ...I want that information available so that I can make an informed choice" (Non-immuniser, #18) Being aware that children could react to the MMR vaccine but not being told what that reaction was or what to expect also caused some hesitation with some mothers. Some mothers hesitated about immunising against Hepatitis B which was at the time of the study recommended for 'at risk' groups. It was unclear to these mothers what this phrase meant and if it applied to their children. [I] believe he should have the hepatitis one 'cause if he comes in contact with another child that's got it, but they say he's not at risk... but what makes him not at risk to get it? So I've been umming and aahing whether to get that one."(Complete Immuniser, #10) The reverse of hesitating to act because of insufficient information was shown by others who figuratively 'shut their eyes' to the information about vaccine risk because it was unsettling. "I think honestly speaking, this sounds stupid, but I think well, I don't want to hear it [about side effects], because it scares me. I know it might be stupid because you think, well you know they're s'posed to have it but if you start thinking well, what if you know if this happens and that happens well, then you wont immunise your children, so, there's a risk I s'pose."(Incomplete immuniser, #42) Another response to a perception of insufficient information was anger or outrage. During the interview some mothers apologised for not being better informed about diseases. Others were angry at their lack of knowledge about diseases and vaccines. This anger was not directed at themselves but at unspecified others. Anger was more often expressed by non-immunisers who believed that drug companies and doctors knew vaccines were not safe but kept the information from the public. Whether parents choose not to act, when action may cause harm, was explored. This was done with the use of two statements-describing whether (1) it would be worse to have your child die due to your action (immunise) or (2) it would be worse due to inaction (die from disease) (see Methods section for statements). Both statements presented uncomfortable possibilities to parents. "You can't ... I mean as far as I'm concerned you lose a child you lose it and it's painful either way I like to think that I've done the best I can to protect him from it um and if you know it's because of the injection well to some degree I'm fatalistic. I mean if it's meant to be it's meant to be. There's not much you can do about it but I would rather know that I've taken every precaution I can instead of you know leaving him open and susceptible to these things." LB: Some people say they would vaccinate because they would feel worse if their child died from an illness which they could have prevented. "Possibly I would sit in that category." (Complete immuniser, #1) Most parents, irrespective of the immunisation status of their children identified more with the second statement, with only a few parents identifying with the first: "I'd have to agree with that. I think if you've given birth to a perfect healthy child and then you've introduced foreign substances into their body which has then damaged them in some way, ah, yeah, I don't know, I don't think I could live with myself. Whereas if they've caught the disease that's kind of c'est la vie you know. I mean it's still awful. It's still a great tragedy, especially if you do lose them. But I think that's that. If you talk about metaphysics, I believe in metaphysics and all the rest of it, so I'd sort of say well, they're meant to be here, they're meant to experience it, they're meant to deal with it or not deal with it depending on what they're here for. So I have to take a philosophical approach. It'd be devastating." (Non-immuniser, #32) Opposite to what would be predicted, many of the non-immunisers disagreed with the first statement and were adamant that this did not form part of their reason not to immunise. FATHER: "I think you would be foolish to reach [that conclusion] I mean we're not foolish. I couldn't possibly say that I would be more comfortable with the you know..." MOTHER: "The child dying or at least you know that he died from the disease." FATHER: "Yeah a natural thing rather than induced. Yeah that's where the natural therapy philosophy goes too far...That would never be reason to [not immunise]." MOTHER: "No, for not immunising him. Yes. I can't even relate to it as a distinction." (Non-immunisers #27) The main reasons parents gave for not agreeing with the first statement was the perception that this scenario was unlikely to occur and, irrespective of immunisation status, most parents believed they had done everything they could to prevent disease. Thus, parents used their perceptions of the risks of the outcome of death from vaccine or the risk of getting the disease to explain their choice. "No, well I'd feel, well I think that that's part of the risk, that there is a small risk that your child will have a reaction to the immunisation that's that's minimal compared to the risk of them getting the disease if you don't immunise so I'd always opt to immunise. (Complete immuniser, #5) "I think you've got more, to me I think she's got more of a chance getting something not being vaccinated than, she's a healthy little girl isn't she?" (Complete immuniser,#7) For those who agreed with Statement 1, they based this on their belief that there was a greater risk from vaccines so the first statement was the more likely scenario. Participants' responses to the two hypothetical radio news items, was concordant with the theories that perception of risk may be influenced by an unrealistic optimism about one's own risks or unrealistic perception of control over one's life. The participants generally did not believe that they would be at risk from the flu. They believed themselves to be healthy, not susceptible to flu and that they were strong enough to fight it off: "I tend to think that couldn't happen to me 'cause I'm young and healthy and couldn't possibly die of flu...Its a few cases [dying] and that happens." (Complete immuniser, #13) Although the scenarios were written specifically so that those being interviewed fitted the 'at risk group', the participants did not identify with this group. They believed that the people who suffered serious consequences of flu were different to themselves. When they heard similar items on the news they assumed the people who were badly affected were old, frail, sick, had not been eating well, had poor immune systems, low resistance or were people who did not look after their health. "It says doctors recommend all should be vaccinated especially those who are overworked, stressed and tired. Well of course they would be the ones whose immune systems would not cope." (Non-immuniser, #19) All non-immunisers believed the risk of this hypothetical flu vaccine was higher than the risk of the disease but some immunisers also perceived high risk and little benefit from the vaccine. They related incidents of relatives who had bad experiences after receiving a flu injection. They believed that there were other ways of reducing the risk of flu such as taking supplements, improving lifestyle and avoiding people with the disease. Those who did believe their family to be at risk of this flu already received annual flu injections, with the occurrence of serious illness in these families prompting them to have flu vaccines. Participants were also asked whether knowing someone who had the illness and had been very ill would affect their decision to immunise against this hypothetical flu. Again the response was that it would depend on the state of the friend's health and their habits: whether they were unhealthy or stressed, or careless of their health. While it would be more concerning to hear of a friend who was ill, most did not think it would mean they themselves were more at risk. In the second news item the 'at risk' group was children under five. All parents stated this item of news would be of greater concern to them. Their first action however, would be to seek more information from their health advisers before immunising. The risk of their children becoming ill was more important than the risk to themselves and all believed that their responsibility as parents was to do everything they could to protect their children. This responsibility made it stressful to make decisions for children because 'you can't afford to make the wrong choice', and one can't take risks for one's children where one might take risks for oneself. "...[I'm] not prepared to risk them, I can control my risks." (Non-immuniser, #29) "You've got to take the risk to prevent them getting sick." (Incomplete immuniser, #20) "Take all the risk factors out of it and make sure they have a good life." (Incomplete immuniser, #34) "Because I am so much more protective of their health than mine. I am concerned that they don't have the option of making choices as easily as I do and that is why I would like to make informed choices. And I feel like if I make the wrong choice for yourself and that is something that I wear, I am responsible for it. If I make the wrong choice for them it is more serious." (Non-immuniser, #21) There was however, a reluctance to immunise. Many believed there was an over-reliance on immunisations and antibiotics and that they would only immunise if the disease was widespread or local (Statewide). If it were not widespread it was not worth the risk of preventative medicine. Table 3 summarises the factors found to influence the decision to immunise and corresponding aspects of the explanatory theories used as a theoretical framework for this study. In this table we aim to show how the information from risk perception and decision making under uncertainty allow for a greater explanation or point to more nuanced action. For example, consideration of unfamiliarity with diseases would be part of the Health Belief Models framework of considering perceived severity. However the Health Belief Model does not explain or allow us to understand what people perceive as familiar or unfamiliar. One might think that if parents are not familiar with a disease they may not think it is serious, whereas subjective perception of risk would indicate that the reverse may be operating: unfamiliarity increases people's perception of risk. Similarly, where parents don't think their child is susceptible to the disease (a Health Belief Model construct), the idea of having 'optimistic control' helps to explain why they might think this. In considering the news reports respondents were asked how many would 'several deaths' be to cause them to worry about the risks of the disease. It was difficult for the participants to respond to this question meaningfully and the production of these numbers was somewhat arbitrary; most were not comfortable giving their response, and many could not say. For instance, one couple said they could give an answer if it was needed to meet the research requirements but it would be meaningless. The numbers given varied from only one or two deaths, five in the State, or between one and ten percent. Others gave figures of more than fifty percent of those who got the disease would have to die for them to be concerned. The question was useful, however, because it provoked participants to define the type of information they wanted in order to make sense of reports such as those they had just heard. For instance, they said their response to the number of deaths would depend on how similar to their own circumstances were those who had died. Did they live in Australia, Victoria or developed countries? Were those who were dying, previously Table 3 Factors influencing the decision to immunise drawing on the Health Belief Model, subjective perception risk and risky decision-making theories healthy people or those who were sick and thus, more susceptible? Was it a familiar disease (flu) or rare (Ebola virus)? If it was unfamiliar it was frightening. If it was familiar (flu) they wanted to know the details of who it was who had died or suffered complications. Thus, it was not the statistics that were important for deciding on risk, but the characteristics of those who had the disease and the familiarity or unfamiliarity of the disease. The decision to immunise or not is complex with perceptions of risks of vaccines, diseases and robustness of the child's health to be considered. In this study we have identified and clarified differences in perception between complete, incomplete and non-immunisers and also identified similarities between all mothers with respect to the decision to immunise their young children. While the decision to immunise young children can be understood to some extent in the context of perceptions of severity and susceptibility to disease and benefits and barriers to immunisations, the theories of risk perception and decision-making add a depth of understanding to the differences found between these parents in terms of their perceptions and interpretations of what is risky and what is not. Two aspects in particular appear to be important: the familiarity or unfamiliarity with the disease and perceived control over risks or outcomes. Thus, perceptions of severity of disease are influenced by the unfamiliarity of the disease and/or the perception that these diseases will have uncontrollable outcomes. That is, diseases with which parents are least familiar are perceived as more severe than those with which parents are more familiar and therefore worth taking preventive action. Being a familiar disease contributed to delays in immunisation, or not immunising as these familiar diseases were not considered to be severe. This finding is congruent with other primary studies. For example, Hilton et al [36] reported 'of all the diseases... measles was the one that parents most commonly reported having as a child. ...Indeed their experience of measles often rendered it a less threatening disease.... While parents with no experience of measles entertained the long-term damage it could inflict, those with experiences of it tended to minimise the risks' (p 174, authors' italics). From this it would appear that it is too simplistic to attribute reduction in immunisation uptake to a growing lack of familiarity with diseases because of the success of the immunisation programmes. Understanding people's perceptions of what can and cannot be controlled is important to understanding their behaviour. There is both an aspect of fearing uncontrollable or unknowable outcomes and therefore taking preventive action and an optimistic belief or an illusion that the environment or risks can be controlled. Unlike the non-immunisers in this study and others [21] , immunisers choose immunisation because they believe they have limited control over their children's environment and contacts. Many believed they could control risks to their own health or were willing to 'take the risk' with their own health. Parents were less willing to take risks with their children's health than with their own. This was partly because children were perceived as more susceptible than healthy adults and partly because their children were dependent on them making good decisions. This unwillingness to take risks included being cautious of preventive action as well as cautious about diseases. An important barrier to action was the tension between what is 'natural' and medical intervention. For many mothers there was something 'unnatural' about medical intervention. They held a belief that medical intervention was necessary for important diseases but that it was not safe or necessary to use for all problems. Again, this perception that what is unnatural is more risky is congruent with the studies of subjective risk perception, but could not be predicted from the Health Belief Model. Importantly, the participants believed that the risks of diseases and complications from disease were not equally spread throughout the community. When listening to reports of epidemics, it is not the number of people who are affected but the familiarity or unfamiliarity of the disease and the characteristics of those who had the disease that caused parents to worry about taking preventive action. Poor information or communication creates barriers to immunisation completion which can be understood in terms of the concepts of outrage and ambiguity. Lack of trust and poor communication between providers and parents exacerbated the belief that information was being kept from them. Because they all believed that parents were ultimately responsible for their children, this feeling that information was denied them frustrated and angered them. This study has used qualitative methods to determine if aspects of theories of risk and decision-making can help to explain parents' decisions about immunising their children. Traditionally qualitative research has been associated with the generation, rather than the testing, of hypotheses, however, this denies a major strength of qualitative research which is to examine theories in the light of data. As such it is a method suited to producing understanding and to generating solutions to problems [34] . We believe this method, therefore, is appropriate to provide a greater understanding of complexities of decision making and perceptions of disease and vaccines and a depth of information not available from large scale quantitatively based surveys. The benefits of using the qualitative method described in this study is the large quantity of detailed information it provides, however, using a small non-randomly selected sample can present problems in determining the generalisabilty of the ensuing information. The results of this study confirm, complement and extend the findings of other studies in this area [4, 5, 36, 37] . The data that this paper draws on were collected in the late 1990s and some may wonder at their currency. We believe that this might be a problem if the focus of the paper was about which diseases or vaccines are an issue; with scares and controversies these can change over time. The point of this paper however, has been to examine the utility of synthesizing theories of health protective behaviours, risk perception and decision-making. While we recognise that different socio-temporal contexts may create different issues (e.g. the impact of the MMR controversy was substantially greater in the UK than Australia), we argue that the approach we have taken in this paper provides a framework for us to make sense of people's reactions to and perceptions of old (and new) diseases and vaccinations at any time. We therefore think this work can contribute to, and be of particular importance in informing the public health approaches to new flu epidemics. It provides data supporting commentary and critique of the current public health approach to the issues of vaccine risk and immunisation uptake, being that continued provision of better risk information is not the answer [7] . We have found and would argue that the theories of risk perception and aspects of decision-making under uncertainty have been useful for understanding the differences and similarities between pro-immunisers and non-immunisers, except for the issue of omission bias. Parents in this study, whether immunisers or not, generally did not agree that a negative outcome was preferable or more acceptable from inaction. This finding raises some doubts about the methods and/or generalisability of findings from studies of this phenomenon, which usually involve multiple, similar, hypothetical situations with limited contextual information, presented in mathematical and probabilistic language. Participants in this study responded to these omission bias statements by generally denying that either contributed to their decision to immunise their child. Similar findings to this study have been reported by others [3] [4] [5] 37] and have important implications for how public health addresses the issues of trust and communicates risk information. As others have noted there is more to risk communication than providing more facts about risks [3, 4, 7, 23] . To paraphrase Hobson-West, from these studies and critiques, it is clear that education is not the main policy tool and ignorance is not the main enemy for maintaining immunisation uptake (p 279 [23] ). Drawing on the aspects of subjective perceptions of risk and decision-making under uncertainty, we believe the following needs to be considered in communicating risk information and health messages: Facts and figures are not interpreted or acted upon rationally: dread, catastrophic potential and familiarity with the risk influences interpretation and action People act as lay epidemiologists. Thus providing information about risks as though everyone has the same risk makes the advice unbelievable and can be discounted Parents are more willing to take risks about their own health than with their children's health but this greater caution about their children's health does not automatically mean they will accept medical intervention From the theory of subjective perception of risk people may well be wary of novel vaccines or therapies (manmade versus natural risks) hence there may be some hesitation in the uptake of such vaccines People will discount their risks-'the people affected are not like me'. This may have implications for how people will assess their risk of being badly affected by any outbreak of new strains of influenza such as H1N1. Clear communication which involves listening to, and not dismissing people's concerns, is valued. This study has shown that there is a need to understand and take into account how people subjectively perceive risks and how that influences their decision-making, in order to understand their choices and behaviours. Theories of risk perception and decision-making can help to explain differences in perceptions of severity and susceptibility of diseases and vaccines. Importantly, this study has found that health messages about the risks of disease which are communicated as though there is equality of risk in the population may be unproductive as these messages are perceived as unbelievable or irrelevant. The findings from this study have implications beyond the issue of childhood vaccinations as we grapple with communicating risks of new epidemics. Using and developing a more complex theoretical approach to public health issues may increase the likelihood we can understand the barriers to action and develop effective methods of communicating risk and delivering acceptable public health interventions. And indeed may usefully contribute to the current debates, especially in the UK, of how these theories of risk and decision-making can be used to 'nudge' other health behaviours [38, 39] .

Development and Applications of VSV Vectors Based on Cell Tropism

Viral vectors have been available in various fields such as medical and biological research or gene therapy applications. Targeting vectors pseudotyped with distinct viral envelope proteins that influence cell tropism and transfection efficiency are useful tools not only for examining entry mechanisms or cell tropisms but also for vaccine vector development. Vesicular stomatitis virus (VSV) is an excellent candidate for development as a pseudotype vector. A recombinant VSV lacking its own envelope (G) gene has been used to produce a pseudotype or recombinant VSV possessing the envelope proteins of heterologous viruses. These viruses possess a reporter gene instead of a VSV G gene in their genome, and therefore it is easy to evaluate their infectivity in the study of viral entry, including identification of viral receptors. Furthermore, advantage can be taken of a property of the pseudotype VSV, which is competence for single-round infection, in handling many different viruses that are either difficult to amplify in cultured cells or animals or that require specialized containment facilities. Here we describe procedures for producing pseudotype or recombinant VSVs and a few of the more prominent examples from envelope viruses, such as hepatitis C virus, Japanese encephalitis virus, baculovirus, and hemorrhagic fever viruses.

Viruses are obligate parasites of living organisms, and their replication is absolutely dependent on the host cell's machinery. The entry of enveloped viruses requires host cell binding and membrane fusion that is mediated by envelope proteins located on the surface of the virion. Some viruses utilize a single molecule as a receptor for entry into the host cell, while many viruses require co-receptor(s) localized near the receptor for complete entry. Identification of viral entry receptors that are composed of membrane proteins, lipids, or carbohydrates is important for examining the life cycle of a virus and for further developing entry inhibitors. However, receptors or co-receptors of several viruses have been difficult to identify because of the lack of reliable cell culture systems, an insufficient amount of native viral particles, or difficulty in handling because of the requirement for biosafety level (BSL)-3 or -4 containment. Therefore, several surrogate systems have been developed to study the initial step of infection. One of the most primitive assays is a binding assay. Purified soluble envelope proteins, viral-like particles which are produced in insect cells by baculoviral vectors, and authentic viral particles obtained from patients have been used to study the mechanisms of viral attachment and to identify binding receptor molecules. However, these binding assays cannot be used to analyze further steps of infection such as fusion and penetration. A cell fusion assay was established to examine the membrane fusion activity of viral envelope proteins. This assay is sensitive and can easily determine cell fusion using reporter genes. Pseudotype virus systems based on vesicular stomatitis virus (VSV), influenza virus, retroviruses, and lentiviruses have also been established to examine entry mechanisms and to identify putative entry receptors for targeted viruses ( Table 1) . Pseudotype viruses have also been applied in neutralization tests for antibodies and vaccine development ( Table 1) . As for the application of a pseudotype virus system for VSV, a recombinant virus system with a heterologous viral envelope gene together with a reporter gene encoded into its own genome instead of the G gene has also been developed. In this paper, we describe the properties of pseudotype or recombinant VSVs and their application to some enveloped viruses we have studied, such as the hepatitis C virus (HCV), Japanese encephalitis virus (JEV), baculovirus, and hemorrhagic fever viruses. Vesicular stomatitis virus is a non-segmented, negative-stranded RNA virus that belongs to the family Rhabdoviridae, genus Vesiculovirus. VSV infects a broad range of animals, including cattle, horses, and swine. The genome of the virus codes for five major proteins, glycoprotein (G), matrix protein (M), nucleoprotein (N), large protein (L), and phosphoprotein (P). The G protein mediates both viral binding and host cell fusion with the endosomal membrane following endocytosis. The L and P proteins are subunits of the viral RNA-dependent RNA polymerase. The simple structure and rapid high-titer growth of VSV in mammalian and many other cells has made it a useful tool in the fields of cellular and molecular biology and virology, and this was further strengthened with the establishment of the reverse www.frontiersin.org genetics system for VSV. Recombinant VSV in which native envelope G protein is replaced with a foreign reporter gene such as a fluorescent reporter protein, luciferase, or secreted alkaline phosphatase (SEAP) can normally bud from producing cells even in the absence of G protein, and heterologous viral envelope proteins are incorporated into the virion. Previous studies demonstrated that VSV forms a "pseudotype" when a cell is co-infected with VSV and other enveloped viruses (Huang et al., 1974; Witte and Baltimore, 1977) . A pseudotype virus is defined as a viral particle harboring other types of viral envelopes or host cellular proteins with or without its own envelope. By virtue of these characteristics of VSV, pseudotype virus systems, in which VSV G proteins are completely replaced with other types of viral envelope proteins, have been established (Figure 1) . Up to the present, numerous types of pseudotype viruses have been constructed with heterologous viral envelope proteins and used in studies examining the entry of viruses, for identification of novel viral receptors, for development of neutralization tests, and as vaccine vectors ( Table 1 ). In particular, availability of pseudotype viruses has been useful in the study of several high-risk viruses that require high-level containment facilities, i.e., in the handling of BSL-3 or -4 viruses. Infectivity of these pseudotype viruses can be easily and quantitatively evaluated by measurement of the reporter gene activities. A recombinant virus system, which encodes a heterologous viral envelope gene instead of an envelope gene in its own genome, has also been made available by establishment of reverse genetics (Figure 2 ). This recombinant virus is replication-competent both in vitro and in vivo and can contribute to the study of targeted viruses that inefficiently replicate in experimental systems. Although the pseudotype virus is limited to single-step infection and therefore provides a poor model for actual infection processes, the recombinant virus is a far more authentic and powerful tool for investigating targeted viral infection. Currently, this system is applicable to the VSV or other FIGURE 1 | Schematic representation of the production of pseudotype VSV. Producer cells were transfected with an expression plasmid encoding foreign envelope genes and then infected with a VSV G-complemented pseudotype virus (*G-VSVΔG). The pseudotype virus released from the producer cells infected target cells but was not able to produce infectious progeny viruses. several viruses and not to retroviruses or lentiviruses. Recombinant VSV can be produced in various cells without regard to transfection efficiency; on the other hand, recovery of pseudotype VSV as well as pseudotype retroviruses or lentiviruses is restricted to 293T or some other type of cells that exhibit a high competency of transfection. Recombinant VSV could also lead to the induction of cellular and humoral host immunity (Schnell et al., 1996) . Seeded or recombinant VSVs in which the G gene is replaced by a foreign reporter gene such as a fluorescent reporter protein (green fluorescent protein, GFP; red fluorescent protein, RFP; and so on), luciferase, or SEAP or each viral envelope gene were generated as described below. Either 293T or BHK cells were grown to 90% confluence on 35-mm tissue culture plates. The cells were infected with a recombinant vaccinia virus encoding the bacteriophage T7 RNA polymerase (vTF7-3) at a multiplicity of infection (MOI) of 5. After incubation at room temperature for 1 h, the cells were transfected with helper plasmids, pBS-N, pBS-P, pBS-L, and pBS-G, and template plasmids, pVSVΔG-GFP (RFP), pVSVΔG-Luci, pVSVΔG-SEAP, or pVSVΔG-Env using a cationic liposome reagent. After 4 h, the supernatants were Table 1 | Application studies of pseudotype and recombinant VSV. Takada replaced with 10% FBS DMEM, and cells were incubated at 37˚C for 48 h. The supernatants were then filtered through a 0.22-μmpore-size filter to remove vaccinia virus and were applied to 293T or BHK cells that had been transfected with pCAGVSVG 24 h previously. If BHK cells constitutively expressing the bacteriophage T7 RNA polymerase (BHKT7) were utilized, the cells were only transfected with helper plasmids, pIRES-N, pIRES-P, pIRES-L, pIRES-G, and template plasmids using a cationic liposome reagent without the vaccinia virus infection. Recovery of the virus was assessed by examining the cells for the cytopathic effects that are typical of a VSV infection after 24 h. Stock of * G-complemented viruses, i.e.,VSVΔG virus or recombinant viruses transiently bearing VSV G protein on the virion surface, were grown from the single plaque on BHK cells transfected with pCAGVSVG and then stored at −80˚C. The infectious titers of the recovered viruses were determined by a plaque assay. To generate pseudotype virus, 293T, BHK, or some other type of cells that exhibit a high competency of transfection were transfected with a plasmid expressing the envelope protein using a cationic liposome reagent. After 24 h of incubation at 37˚C, cells were infected at an MOI of 0.5 with the * G-VSVΔG-Luci and * G-VSVΔG-SEAP, or 5 with * G-VSVΔG-GFP (RFP). The virus was adsorbed for 2 h at 37˚C and then extensively washed four times or more with serum-free DMEM. After 24 h of incubation at 37˚C, the culture supernatants were collected, centrifuged to remove cell debris, and stored at −80˚C. To generate recombinant virus in various mammalian cells, cells were infected with the * G-complemented VSVΔG-Env at an MOI of 5 for 2 h at 37˚C and then extensively washed four times or more with serum-free DMEM. After 24 h of incubation at 37˚C, the culture supernatants were collected and stored at −80˚C. The infectious titers of the viruses were determined by evaluation of each reporter assay or a focus-forming assay. Further details of the protocol can be found in a recent paper (Whitt, 2010) . Hepatitis C virus has already infected more than 3% of the worldwide population and 80% of those infected develop persistent HCV infection (Cerny and Chisari, 1999; Theodore and Fried, 2000) . Persistent HCV infection often leads to chronic hepatitis, hepatic steatosis, cirrhosis, and hepatocellular carcinoma. Currently, there are still 1.5 million or more HCV carriers in Japan. In past years, anti-hepatitis C therapy has modestly improved; however, a currently available combination therapy, consisting of interferon and the nucleoside analog, ribavirin, shows a sustained response in only less than half of the treated patients. The development of innovative treatment alternatives for patients infected with HCV is urgently required, and a better understanding of the life cycle of HCV should allow us to improve HCV therapies. However, due to the lack of an in vitro cell culture system for the isolation of virus directly from patient sera at present, various surrogate systems such as replicon cells (Lohmann et al., 1999) , pseudotype viruses (Lagging et al., 1998; Matsuura et al., 2001; Bartosch et al., 2003; Hsu et al., 2003; Tani et al., 2007) , or trans-complement particles (Ishii et al., 2008; Steinmann et al., 2008) have been developed to study each step of HCV infection. Although in vitro binding assays using soluble purified envelope proteins or HCV-LPs have identified several candidate receptors for HCV, the final determination of a true entry receptor or coreceptor capable of internalizing HCV may be made using an infection assay. Toward this end, pseudotype virus systems based on VSV and retrovirus or lentivirus have been established and applied to identify entry receptors for HCV. Although it is still unknown how HCV envelope proteins retained in the endoplasmic reticulum (ER) are incorporated into both VSV and retroviruses, which naturally bud from the plasma membrane, significant infectivity of these pseudotype viruses has been exhibited in several human hepatoma cell lines. These infections could be inhibited by treatment with antibodies or soluble proteins against putative receptors or HCV envelope proteins, or by a knockdown of receptor molecules by small interfering RNAs (siRNAs), suggesting that innate HCV infection had occurred. We also successfully generated infectious pseudotype and recombinant VSVs incorporating unmodified HCV envelope proteins in hepatic and non-hepatic human cell lines. These viruses exhibited high infectivity in a human hepatoma cell line, Huh7, which is highly susceptible to infection by cell-cultured HCV (HCVcc). The recombinant virus, but not the pseudotype virus, was able to propagate and form foci only in Huh7 cells. The infection of Huh7 cells with pseudotype and recombinant viruses was inhibited by anti-hCD81 and anti-E2 antibodies and by sera from chronic HCV patients. These viruses, as well as pseudotype retroviruses (HCVpp) or HCVcc, were sensitive to the inhibitors of vacuolar acidification, such as ammonium chloride, concanamycin A, or bafilomycin A 1 , or formation of clathrin-coated pits, chlorpromazine, suggesting that these viruses enter via pH-dependent and clathrin-mediated endocytosis into target cells (Blanchard et al., 2006; Tani et al., 2007) . The infectivity of the recombinant virus was inhibited by an ER αglucosidase inhibitor, N -(n-nonyl) deoxynojirimycin (Nn-DNJ), but not by a Golgi mannosidase inhibitor, deoxymannojirimycin www.frontiersin.org (Tani et al., 2007) . Focus formation of the recombinant virus was also impaired by Nn-DNJ treatment. It was obvious that the appearance of infectious or non-infectious viruses was dependent on the cell type as a result of the infectivity of the recombinant viruses generated from various cell lines. Although the precise mechanisms of HCV assembly or budding that cause the differences in infectivity of viruses generated from different cell lines is still unclear, host cellular factors might be involved in the assembly or budding steps in the generation of infectious particles. Japanese encephalitis virus, a mosquito-borne zoonotic pathogen, is the leading cause of viral encephalitis in humans, with ∼50,000 cases reported annually worldwide. JEV is an enveloped virus belonging to the family Flaviviridae and the genus Flavivirus, which also includes Dengue virus, West Nile virus, Yellow fever virus, and Tick-borne encephalitis virus (Gubler et al., 2007) . The genome consists of a single-stranded positive-sense RNA of approximately 11 kb, encoding a single large polyprotein, which is cleaved by host-and virus-encoded proteases into three structural (C, PrM, and E) and non-structural (NS1, NS2A, NS2B, NS3, NS4A, NS4B, and NS5) proteins. The envelope protein (E) is a 53-kDa glycoprotein, which is a major component of the virion surface and has been found to be associated with all the biological properties of the virus, such as attachment to cellular receptors, penetration, fusion with the endosomal membrane, host cell range and cell tropism, and neutralization to antibodies. Although a number of cellular components that interacted with E protein, such as heat shock cognate protein 70 (Ren et al., 2007) , heat shock protein 70 (Das et al., 2009) , vimentin (Das et al., 2011) , glycosaminoglycans (Su et al., 2001; Lee and Lobigs, 2002) , and laminin (Boonsanay and Smith, 2007) , have been shown to participate in JEV binding or penetration, the precise mechanisms remain largely unknown. The pseudotype and recombinant VSV systems have offered us useful tools to focus on the study of entry mechanisms of JEV E proteins by using control viruses harboring an appropriate protein on identical particles. Both pseudotype (JEVpv) and recombinant (JEVrv) VSV bearing the JEV E protein exhibited high infectivity for the target cells, and JEVrv, but not JEVpv, was able to propagate and form foci, as did authentic JEV . Both JEVpv and JEVrv were neutralized by anti-JEV E antibodies. Treatment of cells with inhibitors for vacuolar ATPase and clathrin-mediated endocytosis reduced the infectivity of JEVpv, suggesting that JEVpv enters cells via pH-and clathrin-dependent endocytic pathways. Treatment of the JEVpv and JEVrv with cholesterol drastically reduced the infectivity, as previously reported on authentic JEV (Lee et al., 2008) . In contrast, depletion of cholesterol from the viruses by treatment with methyl β-cyclodextrin enhanced the infectivity. Furthermore, treatment of cells with sphingomyelinase (SMase), which hydrolyzes membrane-bound sphingomyelin to ceramide, drastically enhanced infection with JEVpv and JEVrv . These enhancements were inhibited by treatment with an SMase inhibitor or C 6 -ceramide. Involvement of ceramide in the entry of JEV was confirmed by co-precipitation of the JEV E protein with labeled-ceramide . In our study, it was demonstrated that cellular lipid components such as cholesterol and ceramide play crucial roles in the entry of JEV. Modification of sphingolipids on the plasma membrane of the cells might be a novel target for the development of antivirals against JEV infection. Baculovirus vectors have been shown to exhibit not only a highlevel of gene expression in insect cells but also efficient gene transduction into a wide variety of mammalian cells with lower cytotoxicity (Hofmann et al., 1995; Boyce and Bucher, 1996; Shoji et al., 1997) . In contrast, the complement systems of animals have been defined to represent a potent primary barrier to in vivo application of baculovirus vectors produced in insect cells (Hofmann and Strauss, 1998; Tani et al., 2001 Tani et al., , 2003 . However, pseudotype viruses based on retroviruses or lentiviruses bearing baculovirus envelope protein GP64 have recently been shown to exhibit efficient gene transduction into mouse organs for long periods compared with those bearing the VSV G protein, which is commonly used for pseudotyping (Kumar et al., 2003; Schauber et al., 2004; Kang et al., 2005; Sinn et al., 2005 Sinn et al., , 2008 . It was considered that serum resistance of the pseudotype viruses bearing GP64 protein is caused by differences between insect and mammalian cells, because pseudotype retroviruses or lentiviruses were generated from mammalian cells, and in contrast, baculovirus was generated from insect cells. Therefore, we generated recombinant VSV bearing GP64 protein in both insect and mammalian cells and examined the role of the GP64 on resistance to inactivation by human or guinea-pig sera . Recombinant VSVs generated in human cell lines exhibited the incorporation of human decay accelerating factor (DAF) in virions and were resistant to serum inactivation, whereas those generated in insect cell lines exhibited no incorporation of human DAF and were sensitive to complement inactivation. Recombinant baculoviruses generated in insect cells expressing human DAF or carrying the human DAF gene exhibited resistance to complement inactivation, suggesting that acquisition of resistance to human complement by the incorporation of DAF with baculovirus GP64 represents a step in the development of novel viral vectors for improved gene therapy. in the cells treated with various entry inhibitors depended on the species of pseudotype virus, suggesting that several entry mechanisms were involved in the infection of arenaviruses (Tani et al., unpublished data) . In studies on serological diagnosis of arenaviruses, cross reaction occurred among species of arenaviruses in enzyme-linked immunosorbent assay or immunofluorescence assay, whereas neutralization tests using pseudotype viruses exhibited a specific reaction with each species of virus (Nakauchi et al., 2009; Iha et al., unpublished data) . Although Old or New World arenaviruses have been shown to utilize α-dystroglycan or human transferrin receptor 1, respectively, as one of the cellular receptors, infectivities of the pseudotype viruses have not been consistent with the expression levels of the receptor molecules in our preliminary studies. The infection of pseudotype viruses was not completely inhibited by soluble protein or antibodies of receptor molecules, suggesting that another receptor molecule(s) might be involved in the entry of these viruses. Although further characterization of the pseudotype viruses bearing GPC envelope proteins of arenaviruses will be needed, these viruses are thought to mimic the functional properties of wild type arenaviruses and are suitable for the study of entry mechanisms, including investigation of novel cellular receptor(s), neutralization tests, or vaccine development. Up to the present, various viral vectors aimed at gene transfer or therapy have been developed and applied in biological and medical research fields. Pseudotype or recombinant VSV are useful tools as alternative viruses to study entry mechanisms, identification of novel cellular receptors, screening antiviral libraries, or development of serological diagnosis for various kinds of viruses, especially unmanageable BSL-3 or -4 viruses. These viruses have also been applied in targeting vectors to specific cells. VSV vectors with monoclonal antibodies against specific oncogenic proteins or viral receptor molecule(s) incorporated on virion surface have been targeted specifically to cells expressing oncogenic proteins or infected cells expressing the viral envelope proteins, respectively, without any influences on normal or uninfected cells. This raises the possibility of the elimination of cancer cells or chronic viral infections by using acute VSV infection. Genetically engineered VSVs encoding suicide cassettes or immune response genes have also been generated as more specific, safer, and effective agents for cancer therapies. Further studies and applications of VSV vectors will provide us not only with useful tools for virological studies but also various benefits for biological sciences and medical research. This work was supported in part by grants-in-aid from the Ministry of Health, Labour and Welfare; the Ministry of Education, Culture, Sports, Science, and Technology; the 21st Century Center of Excellence Program of Japan; the Global Center of Excellence Program; and the Foundation for Biomedical Research and Innovation, Japan.

Distinct Regulation of Host Responses by ERK and JNK MAP Kinases in Swine Macrophages Infected with Pandemic (H1N1) 2009 Influenza Virus

Swine influenza is an acute respiratory disease in pigs caused by swine influenza virus (SIV). Highly virulent SIV strains cause mortality of up to 10%. Importantly, pigs have long been considered “mixing vessels” that generate novel influenza viruses with pandemic potential, a constant threat to public health. Since its emergence in 2009 and subsequent pandemic spread, the pandemic (H1N1) 2009 (H1N1pdm) has been detected in pig farms, creating the risk of generating new reassortants and their possible infection of humans. Pathogenesis in SIV or H1N1pdm-infected pigs remains poorly characterized. Proinflammatory and antiviral cytokine responses are considered correlated with the intensity of clinical signs, and swine macrophages are found to be indispensible in effective clearance of SIV from pig lungs. In this study, we report a unique pattern of cytokine responses in swine macrophages infected with H1N1pdm. The roles of mitogen-activated protein (MAP) kinases in the regulation of the host responses were examined. We found that proinflammatory cytokines IL-6, IL-8, IL-10, and TNF-α were significantly induced and their induction was ERK1/2-dependent. IFN-β and IFN-inducible antiviral Mx and 2′5′-OAS were sharply induced, but the inductions were effectively abolished when ERK1/2 was inhibited. Induction of CCL5 (RANTES) was completely inhibited by inhibitors of ERK1/2 and JNK1/2, which appeared also to regulate FasL and TNF-α, critical for apoptosis in pig macrophages. We found that NFκB was activated in H1N1pdm-infected cells, but the activation was suppressed when ERK1/2 was inhibited, indicating there is cross-talk between MAP kinase and NFκB responses in pig macrophages. Our data suggest that MAP kinase may activate NFκB through the induction of RIG-1, which leads to the induction of IFN-β in swine macrophages. Understanding host responses and their underlying mechanisms may help identify venues for effective control of SIV and assist in prevention of future influenza pandemics.

Swine influenza is an acute respiratory disease caused by swine influenza viruses (SIV). The symptoms and signs generally include fever, sneezing, nasal rattles, and respiratory distress in pigs. Pigs recover within a few days, but severe signs can develop and mortality can reach up to 10% when highly virulent strains are involved [1] or pigs are infected at young ages [2, 3] . Pigs have long been considered to be the intermediate host of various subtype viruses and ''mixing vessels'' for the evolution and genesis of influenza viruses with pandemic potential because of their susceptibility to swine, avian, and human influenza viruses [4, 5, 6] . This broad susceptibility is due to the presence of both sialic acid (SA)2,3 Gal-and SA2,6-Gal receptors present in the respiratory epithelium. Three major SIV subtypes are prevalent: H1N1 (classical swine H1N1 and avian-like H1N1), H3N2 (triple reassortant H3N2 and human-like H3N2), and H1N2 [2, 7, 8, 9, 10, 11] . Pigs are also susceptible to and show clinical signs when infected with pandemic (H1N1) 2009 virus (referred to hereafter as H1N1pdm) [12] , which emerged in April 2009 in North America [13] , arising at least in part from contemporaneous SIV. To date H1N1pdm has been found in a few swine farms [12, 14, 15] , which further demonstrates a two-way process of both gene and virus trafficking between humans and pigs. Though H1N1pdm has remained antigenically and genetically stable in humans since its emergence, a novel reassortant SIV containing a H1N1pdm-like NA and seven other genes from triple-reassortant H1N2 and European ''avian-like'' H1N1 viruses was identified in early 2010 [16] , and that same year H1N1pdm was shown to be evolving genetically at a faster pace in pigs than it was in humans [12, 15, 17] . Effective control of circulating influenza viruses in swine populations is key to reducing consequent genesis of novel pandemic strains that threaten the health of both humans and animals. Studies have been conducted to identify proinflammatory cytokines including TNF-a, IL-6, IL-12, and IFN-a or IFN-c, which are upregulated in lung or bronchoalveolar secretions in SIV-infected pigs [18, 19, 20, 21] and may be correlated with clinical manifestations. In an alveoli macrophage-depleted pig model, macrophages appeared to be indispensible to effective clearance of SIV from lungs. A higher frequency of cytotoxic T, cd T, and Treg cells were also detected in infected pig lungs [18] , which together with the induction of cytokines, contribute to pathogenesis of influenza infection in pigs. Exploring the mechanism of regulation of host responses is crucial for understanding the pathogenesis of SIV and for controlling swine influenza in pigs. Macrophages residing beneath the respiratory epithelium and surrounding alveoli are part of the first line defenses against influenza viruses. During influenza viral replication in bronchial epithelial cells, macrophages are one of the earliest targets to be infected. Together with dendritic cells, macrophages coordinate innate immune responses, which subsequently lead to adaptive immunity by initiating antigen presentation and lymphocyte activation. Macrophages are indispensable in alveolar host defense and controlling influenza virus in pulmonary organs in pigs [22] . While protective in launching host antiviral responses and restricting virus spread, induced proinflammatory cytokines and chemokines are also the cause of pathogenicity for the host and may lead to acute respiratory failure (ARF), a major cause of death in highly pathogenic H5N1 or H1N1pdm-infected humans [23] . Needless to say, the roles of macrophages are critical to pathogenicity as well as host protection in SIV-infected pigs. However, little is known about the mechanisms of how host responses are regulated in pigs or their macrophages. Considering the critical role macrophages play in SIV infections, and the threat that H1N1pdm could further evolve higher virulence in pigs and subsequently infect humans, we were interested in profiling host responses of swine macrophages to H1N1pdm, and more importantly, in exploring the underlying mechanism of host response regulation including antiviral, proinflammatory responses, and apoptosis in pigs. In this report, we will demonstrate that swine macrophages are susceptible to infection by H1N1pdm. We will show a unique pattern of proinflammatory cytokine responses to the infection, which are distinctly regulated by swine mitogen-activated protein (MAP) kinases. We have also observed cross-talk between MAP kinase and NFkB pathways, and our data indicate that MAP kinase ERK1/2 and JNK1/2 may impact the activation of NFkB through the induction of RIG-1, leading to IFN-b induction in H1N1pdm-infected swine macrophages. The 3D/4 cells used in our study are a spontaneouslytransformed line of swine macrophages purchased from ATCC (Manassas, VA) and grown in RPMI 1640 medium (Invitrogen) containing 10% fetal bovine serum (FBS). Mouse anti-ERK and anti-JNK antibodies as well as rabbit anti-phospho ERK and antiphospho JNK antibodies (Cell Signaling), anti-cytochrome c, anti-influenza NS1, and alkaline phosphatase (AP)-conjugated anti-rabbit and anti-mouse IgG antibodies (Santa Cruz Biotechnology) were obtained from their respective providers. Anticleaved caspase antibody was obtained from Cell Signaling Technology, and anti-Bak antibody was obtained from EMD Chemicals. The chemicals purchased from EMD Chemicals also included inhibitors for MAP kinases, U0126 (ERK1/2), SB203580 (p38), and InSolution JNK Inhibitor II (JNK1/2), and the inhibitors for NFkB and IKK (6-Amino-4-(4-phenoxyphenylethylamino) quinazoline (Cat. 481406) and Wedelolactone (Cat. 401474), respectively). A/Nanjing/108/2009 (H1N1), a pandemic (H1N1) 2009 virus, was isolated from a swab sample of an outpatient febrile child at the Nanjing Children's Hospital during the pandemic in 2009, Nanjing, China. The sampling procedure was performed in accordance with the rules set by the Institutional Review Board at the Hospital. The eight genomic segments of this virus have been fully sequenced and the raw data are deposited at Genbank under accession numbers JQ173100 through JQ173107. The virus was grown in 9-day-old embryonating chicken eggs; virus allantoic fluid (VAF) was harvested 48 hrs after inoculation, then titrated with standard haemagglutination tests (HA) and plaque assays in MDCK cells for HA and infectious viral titers, respectively [24] . For viral infection, the 3D/4 cells were trypsinized, resuspended in RPMI 1640 medium containing 10% FBS, and plated on 6-cm tissue culture plates at 5610 6 cells per plate 12 hrs before infection. The cells were infected with H1N1pdm inocula in VAF at a multiplicity of infection (MOI) of 1. After 1 hr of adsorption, the virus inocula were discarded and 3 ml of serum-free RPMI 1640 medium containing TPCK-trypsin (1 mg/ml, Sigma) was added. The cells were incubated at 37uC and 5% CO 2 for various time points before cell lysates or total RNA extraction were prepared. mRNA transcript levels of IFN-b, IL-1b, IL-6, IL-8, CCR5, IP-10, TNF-a, FasL, TRAIL, Mx, 2959-OAS, retinoic acid-inducible gene I (RIG-1), melanoma differentiation-associated antigen 5 (MDA-5), and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) genes were analyzed by a two-step real-time RT-PCR assay as described previously [25] . 1 mg of total RNA, prepared from the cells using the RNeasy kit (Qiagen), was reverse transcribed with the QuantiTect reverse transcription kit (Qiagen) following the manufacturer's instructions. The sequences of primers used in the study are listed in Table 1 . The RT reaction was carried out with the RNA after treatment with DNase I at 42uC for 2 min. Real-time PCR was conducted with 1 ml of cDNA in a total volume of 25 ml with the iQ SYBR Green Supermix (Bio-Rad) following the manufacturer's instructions. Relative expression values were normalized using an internal GAPDH control. The fold change of relative gene expression levels was calculated following the formula: 2 (DCt of gene2DCt of GAPDH) [25, 26] . For each reaction, melting curves were analyzed to determine the specificity of each amplicon. To determine the viral RNA level, the total RNA from infected cells was reverse transcribed and cDNA used for Taqman-based real-time PCR (Applied Biosystems) to measure viral M gene transcripts in the infected cells [27] . Cell lysates were prepared by lysing uninfected and infected 3D/4 cells in 1% NP-40 lysis buffer containing 1 mM PMSF, 1% aprotinin, 20 mg leupeptin ul 21 and 1 mM sodium vanadate (Sigma) as described previously [10] . Cell lystes were clarified by low speed centrifugation (1000 g, 5 min at 4uC) and subjected to SDS-PAGE (10 to 12%). Proteins were transferred to the Immuno-Blot PVDF membrane (Bio-Rad), and western blot analysis was performed following standard protocols [28] using rabbit or mouse anti-MAP kinase or phosphor-MAP kinase antibodies (1:500) in TBST containing 5% fat-free milk powder for 90 mins incubation at RT. After washes, incubation with APconjugated anti-rabbit or anti-mouse IgG antibody for another 90 mins followed. After incubation and thorough washes, BCIP/ NBT reagents (Sigma) were used for the development of colorimetric signals on the membrane. The membrane was also blotted with a monoclonal anti-actin antibody (Santa Cruz Biotechnology) for input control. For statistical analysis, a two-tailed Student's t-test was used to evaluate realtime RT-PCR data. An x 2 analysis was used to evaluate significant differences of the data in two and more groups. The 0.05 level of probability (p,0.05) was considered statistically significant. To examine the susceptibility of pig macrophages to H1N1pdm originating from a human host, we infected 3D/4 cells with the A/ Nanjing/108/2009 (H1N1). Typical cytopathic effect (CPE) appeared 16 hrs post infection and the cell monolayer was destroyed 32 hrs post infection (Fig. 1A) . This result demonstrated that H1N1pdm retains the ability to infect and replicate in swine macrophages, and can reach 1.8610 4 PFU/ml as shown in a replicative curve (Fig. 1B) . Apoptosis occurred and proceeded through the course of the infection, as we observed cleaved/ activated caspase-9 as well as the emergence of downstream executioner caspases-6, -7, and -3, which eventually destroyed the infected swine macrophages (Fig. 1C) . Clearly, cytochrome c was released into the cytosol (Fig. 1E) , which activated mitochondriamediated intrinsic apoptosis as early as 3 hrs post infection. Bak, a pro-apoptotic Bcl-2 family member, was upregulated as detected in the infected cells (Fig. 1D) , and may be involved in the release of cytochrome c from mitochondria in swine macrophages. To elucidate the pathogenesis of H1N1pdm in pigs, we examined the pattern of cytokine responses in pH1N1-infected swine macrophages. Total RNA from infected and uninfected 3D/ 4 cells collected at different time points post infection (p.i.) were prepared and used for realtime RT-PCR analyses with specific primers to swine cytokines. We found that the levels of proinflammatory cytokines IL-6 and IL-8 were upregulated up to 51-and 38-fold at 16 hrs, respectively, and the level of IL-8 continued to rise up to 142-fold at 36 hrs p.i.. However, the level of IL-1b remained unchanged throughout the infection ( Fig. 2A) , indicating that IL-6 and IL-8, as well as TNF-a (Fig. 2B ) as described later, were the main proinflammatory cytokines upregulated. We observed a robust induction of antiviral IFN-b, which rose up to 620-and 5,100-fold at 16 and 36 hrs p.i., respectively (Fig. 2C ). IFN-inducible antiviral proteins Mx and 295.-OAS were induced accordingly up to 910-and 12,510-fold, respectively, at 36 hrs p.i. (Fig. 2C) . TNF family members were also induced in response to H1N1pdm infection, which may be attributable to cell death. We found that in pig macrophages the levels of FasL and TNF-a remained undetectable, while TNF-related apoptosis-inducing ligand (TRAIL) seemed to be most abundant before infection, based on Ct values from realtime RT-PCR (data not shown). FasL and TNF-a were induced most robustly, but TRAIL was only mildly induced in response to infection (Fig. 2B) . Among the induced, the level of TNF-a, critical in both cell death and inflammation, was sharply upregulated up to 14-and 162-fold, and FasL up to 43-and 22-fold at 16 and 36 hrs p.i., respectively. FasL and TNF-a may play a major role in H1N1pdm-triggered extrinsic apoptosis. To understand the mechanism of proinflammatory cytokine and TNF family ligand induction in H1N1pdm-infected swine macrophages, we investigated how MAP kinases were activated and whether their signaling pathways were involved in the regulation of various cytokines and TNF family ligands in pig immune cells. 3D/4 cells were infected with H1N1pdm, and cell lysates were prepared at various time points for SDS-PAGE and western blot analyses with specific anti-ERK1/2 and anti-JNK1/2 antibodies. Activated forms of ERK and JNK (phospho-ERK1/2 and phosphor-JNK1/2) were detected by anti-phospho-ERK1/2 and anti-phospho-JNK antibodies. As shown in Figure 3A , ERK1/2 was basally phosphorylated at a low level before infection, but further phosphorylated between 9 and 18 hrs and thereafter p.i.. Phosphorylation and activation of JNK1/2 appeared at 9 hrs and increased to the peak around 18 hrs p.i. (Fig. 3B) . Although both ERK1/2 and JNK1/2 were activated in response to H1N1pdm infection in swine macrophages, ERK1/2 remained active at basal level even before infection, so did JNK1/2 as shown in some of our experiments (Fig. 4B) . However, our data showed that basal level phosphorylation of both ERK1/2 and JNK1/2 remained unchanged in uninfected 3D/4 cells through the period of our infection. In addition to ERK1/2 and JNK1/2, we have also observed the phosphorylation and activation of p38 MAP kinase in H1N1pdm-infected cells (data not shown). To evaluate the role of MAP kinases in the regulation of proinflammatory cytokine responses in H1N1pdm-infected swine macrophages, we pre-treated 3D/4 cells with specific inhibitors for ERK1/2, p38, and JNK1/2 1 hr prior to infection. We then infected the cells with the virus and observed how infectioninduced activation of MAP kinases was affected by inhibition of the respective MAP kinases. As shown in Figure 4A , 3D/4 cells were pre-treated with inhibitors of ERK1/2 (U0126), p38 (SB230058), and JNK1/2 (JNK InSolution), at concentrations of 10 mM, 5 mM, and 50 mM, respectively. While the phosphorylation of ERK1/2 was unaffected by treatment with the p38 and JNK inhibitors, it was completely abolished at both 18 and 30 hrs p.i. (lines 4-5) by the ERK1/2 inhibitor U0126 (Fig. 4A) . We noted that the basal level phosphorylation of ERK1/2 diminished in the presence of U0126. On the other hand, in light of the p38 and JNK inhibition with their specific inhibitors, the phosphorylation of ERK1/2 appeared to be enhanced (Fig. 4A, lines 6-9 ), indicating that a compensatory mechanism may exist among MAP kinases. We observed a similar response in which a complete suppression of JNK1/2 phosphorylation was observed (lines 8-9) when the cells were pre-treated with the JNK1/2 inhibitor (Fig. 4B) . However, the phosphorylation of JNK1/2 was not suppressed at all by the inhibitors of ERK1/2 and p38. We noted that there were double bands for JNK1, and a lower band of JNK1 usually appeared at a later stage of infection (30 hrs p.i.). This band was detected mainly by anti-JNK1/2, but not by anti-phospho-JNK1/ 2, indicating that JNK activation was transient and dephosphorylation of JNK occurred at later stages of infection, probably by an uncharacterized MAP kinase phosphatase (MKP) present in pigs. A basal level phosphorylation of NFkB was also observed in 3D/4 pig macrophages, and was further enhanced upon H1N1pdm infection, indicating that the NFkB pathway was activated as well in infected pig macrophages (Fig. 4C) . When the cells were pre-treated with specific inhibitors of NFkB (10 nM) or IKK (10 mM), the phosphorylation/activation of NFkB was effectively decreased or diminished. MAP kinases and NFkB pathways were activated in H1N1pdm infected pig macrophages, which could be reversed or inhibited by their specific inhibitors. We used these inhibitors to study the regulation of host responses, which may be controlled by these pathways. To determine how cytokine responses are regulated by individual MAP kinases, we pre-treated the cells with ERK1/2 and JNK1/2 inhibitors, respectively, and measured the induction of the cytokines after infection with realtime RT-PCR. We observed that IL-1b was barely detected and not induced during H1N1pdm infection. Interestingly, we noticed that IL-1b was upregulated in the presence of the JNK inhibitor, although no change was observed after the treatment by the ERK inhibitor, indicating that IL-1b could have been induced in swine macrophages infected with H1N1pdm, but was virtually suppressed by JNK1/2 (Fig. 5A) . We observed that the induction of IL-6, IL-8, and IL-10 was completely suppressed in the presence of the ERK1/2 inhibitor, which indicates that IL-6, IL-8, and IL-10 inductions are all dependent on the ERK signaling pathway (Fig. 5B-D) . It is interesting to note that JNK1/2 may play different roles in the induction of IL-6, IL-8, and IL-10 based on their responses in the presence of the JNK inhibitor. JNK1/2 may have moderate effects in the induction of IL-6 ( Fig. 5B ), but may be not relevant at all to the induction of either IL-8 or IL-10 ( Fig. 5C-D) . We also noted that CCL5 (RANTES) was strongly regulated by ERK1/2 and JNK1/2 in swine immune cells. As shown in Figure 5E , induction of CCL5 was efficiently blocked in the presence of either ERK1/2 or JNK1/2 inhibitors, indicating that CCL5 is induced by H1N1pdm infection through ERK and JNK signaling pathways. As for antiviral IFN-b, which was robustly induced with H1N1pdm infection in swine macrophages, ERK1/2 appeared to be essential since the induction of its mRNA transcripts was virtually abolished in the presence of the ERK inhibitor (Fig. 6A ). JNK1/2 may also play a role in IFN-b induction because of its significant decrease at the earlier stage of infection (16 hrs p.i.) when 3D/4 cells were pre-treated with the JNK inhibitor. However, ERK1/2 seemed to be the primary pathway in the IFN-b induction in swine macrophages. The distinct contributions to the induction of IFN-b by ERK1/2 and JNK1/2 were also reflected in the decreased mRNA transcript levels of IFN-inducible antiviral proteins, Mx and 2959-OAS, in the presence of ERK and JNK inhibitors, respectively (Fig. 6B-C) , which is in accordance with the suppression of the IFN-b induction by these same compounds in the infected cells. Both Mx and 2959-OAS were suppressed significantly by the ERK inhibitor, but only the In contrast to the abundance of TRAIL transcripts, mRNA levels of FasL and TNF-a were barely detectable by realtime RT-PCR in swine macrophages (data not shown). However, both FasL and TNF-a were induced profoundly in response to pH1N1 infection ( Fig. 2B and 7A-C) , while the change of TRAIL was mild. By using inhibitors, we concluded that the induction of FasL and TNF-a are mainly controlled by the ERK1/2 and JNK1/2 pathways in pig macrophages. The NFkB pathway could also be critical in host responses, as has been shown in humans and mice infected with influenza A virus. NFkB can be phosphorylated and activated in swine macrophages in response to H1N1pdm infection ( Fig. 8A and 4C ), albeit at a later stage. Interestingly, when the cells were pre-treated with ERK1/2 or JNK1/2 inhibitors, the phosphorylation of NFkB was also suppressed. However, when the cells were pre-treated with the p38 inhibitor, NFkB phosphorylation decreased much less than with ERK1/2 or JNK1/2 inhibitors (Fig. 8A) . This result suggests that a cross-talk may exist between MAP kinase and NFkB pathways, and that among the MAP kinases, ERK1/2 and JNK1/2 are mainly involved. Figure 5 . Regulation of swine proinflammatory cytokine gene transcripts by MAP kinases. 3D/4 cells were pretreated with U0126 and InSolution JNK inhibitor, which are inhibitors of ERK1/2 and JNK1/2, respectively, 1 hr before H1N1pdm infection. Total RNA was prepared at 24 and 36 hrs post infection for reverse transcription. cDNA was used for realtime PCR with specific primers to measure fold changes of cytokine transcripts at different time points. Each assay was repeated at least twice. A-E. Regulation of IL-1b, IL-6, IL-8, IL-10, and CCL5, respectively, by ERK1/2 and JNK1/ 2 inhibitors. Data show mean fold changes plus standard deviation of two or three independent assays. *p,0.05, Student's t-test. doi:10.1371/journal.pone.0030328.g005 We next examined the expression levels of RIG-1 and MDA-5, the RLR family members and cytosolic sensors for RNA viruses. We found that RIG-1 in particular was significantly induced up to 1280-fold, while MDA-5 was also upregulated up to 42-fold in infected pig macrophages (Fig. 8B) . We further examined the induction of RIG-1 and MDA-5 and their relevance to MAP kinases. To do this, we pre-treated the cells with inhibitors of MAP kinases. As shown in Figure 8C , the induction of RIG-1 was completely abolished by the inhibition of ERK1/2 or JNK1/2 inhibitors, and to a much lesser extent, by the p38 inhibitor, suggesting that the induction of RIG-1 was dependent on ERK1/ 2 and JNK1/2, but not as much on p38. This differentially regulated pattern of RIG-1 induction by ERK1/2, p38, and JNK1/2 was similar to the suppression of NFkB phosphorylation/activation by MAP kinases (Fig. 8A) , suggesting that the induction of RIG-1 was associated with ERK1/2 or JNK1/2 activation, but to a much lesser extent with p38. Since NFkB could be downstream activated by RIG-1/IPS-1 [29, 30] , we postulate that ERK1/2 or JNK1/2 may activate NFkB through the activation of RIG-1/IPS-1 during H1N1pdm infection in pig macrophages. A similar, albeit less dramatic, induction and suppression of MDA-5 expression was also observed (Fig. 8D) , which indicated that MDA-5 might also be an intermediate adaptor bridging the MAP kinases ERK1/2 and JNK1/2 to the NFkB pathway activation. The cells were pretreated with U0126, SB230058, and InSolution JNK inhibitor, 1 hr before H1N1pdm infection. Total RNA was prepared from infected cells for reverse transcription. cDNA was used for realtime PCR with RIG-1 and MDA-5 primers to measure fold changes of RIG-1 and MDA-5 transcripts in treated swine macrophages. Each assay was repeated at least twice. Data show mean fold change plus standard deviation of two or three independent assays. *p,0.05, Student's t-test. doi:10.1371/journal.pone.0030328.g008 In the present study, we have demonstrated a pattern of host responses in swine macrophages to H1N1pdm infection. Strong proinflammtory and antiviral cytokine responses including IL-6, IL-8, TNF-a, as well as IFN-b, were observed. In contrast, IL-1b was not induced, and was barely detectable in pig macrophages. This pattern differs from that in bronchoalveolar secretions of SIV-infected pigs in which IL-1b was induced but IL-8 was not [19, 20, 21, 22, 25] . The different cell types involved (macrophages and epithelial cells) may account for the difference. It has previously been reported that in human immune cells and patients a weak innate immune response, evidenced by a poor induction of proinflammatory and antiviral cytokines including IFN-b and TNF-a, has been observed in human monocyte-derived DCs and macrophages infected with H1N1pdm, compared to seasonal H1N1 infection [31] . Highly pathogenic H5N1 viral infection in human macrophages induced higher expression of IL-6 and CCL5 (RANTES) than pH1N1 [32] , which may explain generally mild clinical disease among H1N1pdm-infected patients. In human macrophages, similar to our findings, IL-1b was not detected. MAP kinase signaling pathways and their roles in the regulation of cytokines and viral replications have not been characterized in influenza-infected pig immune cells. In this study, we found that ERK1/2 and JNK1/2 could both be activated in swine macrophages. We noted that ERK1/2 was phosphorylated and active at a low level constitutively, which may be important for the rapid physiological responses required upon infection. To elucidate the mechanism that regulates swine host responses, we used specific inhibitors of MAP kinases to pre-treat macrophages before infection. We determined that the induction of IFN-b, IL-6, IL-8, and IL-10 were regulated by ERK1/2, while JNK1/2 may only play a minor or no role in the regulation of these cytokines. As described earlier, IL-1b was not induced in response to the pH1N1 infection, which could be explained by our data indicating that its induction was in fact efficiently suppressed by JNK1/2 in swine macrophages. This may be the first time that JNK1/2 inhibitory effects on the induction of proinflammatory cytokines have been demonstrated. Previous studies found that IFN induction was dependent on the JNK1/2 signaling pathway in epithelial cells infected with influenza virus infection [33] . However, our data clearly demonstrate that ERK1/2 plays a major role in the regulation of IFN-b in pig macrophages, which may indicate that the regulation of IFN differs in different cell types. We noted that basal level activities of both ERK1/2 and JNK1/2 were constitutively present in non-infected 3D/4 cells, which may be important in the induction of proinflammatory and antiviral cytokines at the early stages of infection. Our data indicate that the induction of IL-6, IL-8, IL-10, CCL-5, as well as IFN-b, were apparent at the earliest stages of viral infection even before ERK1/2 was further activated. We realized that a transformed monocytic cell line, instead of primary cells, was used in the study, which may compromise the significance of our data. Basal level phosphorylation of both ERK1/ 2 and JNK1/2, which may affect certain cytokine production, would be minimal in primary monocytes. However, specific inhibitors used in the study completely wiped out phosphorylation of both ERK1/2 and JNK1/2 (Fig. 4) . The effect of MAP kinase phosphorylation and activation on the regulation of affected cytokines as observed in our study with the inhibitors is, therefore, valid, even though the cells were not primary cultures. Macrophages appear to die inevitably of apoptosis when infected with influenza virus [26] . The Fas-mediated extrinsic apoptotic pathway is apparently triggered by TNF family ligands. While both FasL and TNF-a were induced vigorously upon the viral infection, induction of TRAIL was rather mild in H1N1pdminfected swine macrophages. We knew previously that FasL and TNF-a were barely detectable, while the level of TRAIL remained high prior to the infection based on our realtime RT-PCR data (Ct) (Xing et al., unpublished data). We can therefore presume that H1N1pdm-induced apoptosis may be mainly attributed to FasL and TNF-a, while pig macrophages could be resistant to TRAIL, since the cells remained intact despite the presence of a high level of TRAIL before infection. Furthermore, we were also able to determine that both ERK1/2 and JNK1/2 were involved in the induction of FasL, TNF-a, and TRAIL. FasL is also regulated by ERK1 in chicken macrophages infected with an H9N2 avian influenza virus [34] . Both toll-like receptors (TLR) and RNA helicases, such as RIG-1 and MDA-5, are critical to antiviral innate immunity [35, 36] . As a cytosolic sensor, RIG-1 binds to dsRNA and viral ssRNA that contain a 59-triphosphate not present in host RNA, and then is recruited to mitochondrial protein IPS via the CARD domain, leading to activation of NFkB, IRF-3/-7, and induction of IFN [37, 38, 39] . RIG-1 can be induced by viral infection [40] . In this study, we observed a robust induction of RIG-1 and MDA-5 in H1N1pdm-infected swine macrophages, which appeared to be suppressed completely by inhibitors of ERK1/2 or JNK1/2, but to be a much lesser extent, by the inhibitor of p38. This indicates that the induction of RIG-1 or MDA-5 depends on the activation of ERK1/2 and JNK1/2 in pig macrophages. We postulate a mechanism, therefore, that the cross-talk between MAP kinase and NFkB pathways is through the regulation of RIG-1 and maybe MDA-5, and that ERK1/2 controls the activation of NFkB, leading to the induction of IFN in swine macrophages.

RNA-Seq Based Transcriptional Map of Bovine Respiratory Disease Pathogen “Histophilus somni 2336”

Genome structural annotation, i.e., identification and demarcation of the boundaries for all the functional elements in a genome (e.g., genes, non-coding RNAs, proteins and regulatory elements), is a prerequisite for systems level analysis. Current genome annotation programs do not identify all of the functional elements of the genome, especially small non-coding RNAs (sRNAs). Whole genome transcriptome analysis is a complementary method to identify “novel” genes, small RNAs, regulatory regions, and operon structures, thus improving the structural annotation in bacteria. In particular, the identification of non-coding RNAs has revealed their widespread occurrence and functional importance in gene regulation, stress and virulence. However, very little is known about non-coding transcripts in Histophilus somni, one of the causative agents of Bovine Respiratory Disease (BRD) as well as bovine infertility, abortion, septicemia, arthritis, myocarditis, and thrombotic meningoencephalitis. In this study, we report a single nucleotide resolution transcriptome map of H. somni strain 2336 using RNA-Seq method. The RNA-Seq based transcriptome map identified 94 sRNAs in the H. somni genome of which 82 sRNAs were never predicted or reported in earlier studies. We also identified 38 novel potential protein coding open reading frames that were absent in the current genome annotation. The transcriptome map allowed the identification of 278 operon (total 730 genes) structures in the genome. When compared with the genome sequence of a non-virulent strain 129Pt, a disproportionate number of sRNAs (∼30%) were located in genomic region unique to strain 2336 (∼18% of the total genome). This observation suggests that a number of the newly identified sRNAs in strain 2336 may be involved in strain-specific adaptations.

Systems biology approaches are designed to facilitate the study of complex interactions among genes, proteins, and other genomic elements [1, 2, 3] . In the context of infectious disease, systems biology has the potential to complement reductionist approaches to resolve the complex interactions between host and pathogen that determine disease outcome. However, a prerequisite for systems biology is the description of the system's components. Therefore, genome structural annotation or the identification and demarcation of boundaries of functional elements in a genome (e.g., genes, non-coding RNAs, proteins, and regulatory elements) are critical elements in infectious disease systems biology. Bovine Respiratory Disease (BRD) costs the cattle industry in the United States as much as $3 billion annually [4, 5] . BRD is the outcome of complex interactions among host, environment, bacterial, and viral pathogens [6] . Histophilus somni, a gramnegative, pleomorphic species, is one of the important causative agents of BRD [6] . H. somni causes bovine infertility, abortion, septicemia, arthritis, myocarditis, and thrombotic meningoencephalitis [7] . H. somni strain 2336, the serotype used in this study and isolated from pneumonic calf lung, has a 2.2 Mbp genome and 2044 predicted open reading frames (ORFs), of which 1569 (76%) have an assigned biological function. Genome structural annotation is a multi-level process that includes prediction of coding genes, pseudogenes, promoter regions, repeat elements, regulatory elements in intergenic regions such as small non-coding RNAs (sRNA), and other genomic features of biological significance. Computational gene prediction methods such as Glimmer [8] or GenMark [9] use Hidden Markov models which are based on a training set of well annotated genes. Although these methods are quite efficient, they often miss genes with anomalous nucleotide composition and have several well-described shortcomings: because bacterial genomes do not have introns, detecting gene boundaries is comparatively difficult; due to the usage of more than one start codon, computational genome annotation methods may predict overlapping ORFs [10] ; prediction programs use arbitrary minimum cutoff lengths to filter short ORFs, which may lead to under-representation of small genes. In case of sRNA (small non-coding RNA) prediction, the lack of DNA sequence conservation, lack of a protein coding frame, and the limited accuracy of transcriptional signal prediction programs (promoter/Rho terminator prediction) confound computational prediction [11, 12] . Computational prediction methods are a ''first pass'' genome structural annotation. Whole genome transcriptome studies (such as whole genome tiling arrays [13, 14, 15] and high throughput sequencing [16, 17] ) are complementary experimental approaches for bacterial genome annotation and can identify ''novel'' genes, gene boundaries, regulatory regions, intergenic regions, and operon structures. For example, a transcriptomic analysis of Mycoplasma pneumoniae identified 117 previously unknown transcripts, many of which were non-coding RNAs, and two novel genes [18] . Transcriptome analyses identified novel, non-coding regions in other species, including 27 sRNAs in Caulobacter crescentus [15] , 64 sRNAs in Salmonella Typhimurium [17] , and a large number of putative sRNAs in Vibrio cholerae [16] . sRNAs found in pathogen genomes are known to be involved in various housekeeping activities and virulence [19] . In this study we used RNA-Seq for the experimental annotation of the H. somni strain 2336 genome and to construct a single nucleotide resolution transcriptome map. Novel expressed elements were identified, and where appropriate, computational predictions of previously described gene boundaries were corrected. In 2008 the complete genome sequence of the H. somni strain 2336 became available (GenBank CP000947). The 2,263,857 bp circular genome has a GC content of 37.4%, and 87% of the sequence is annotated to coding regions. The genome has 2065 computationally predicted genes, of which 1980 are protein coding. We sequenced the transcriptome of H. somni using Illumina RNA-Seq methodology, and obtained 9,015,318 reads, with an average read length of approximately 76 bp. We mapped approximately 9.4% reads onto the reference DNA sequence of H. somni strain 2336 using the alignment program Bowtie [20] . To determine expressed regions in the genome, we estimated the average coverage depth of reads mapped per nucleotide/base. We used pileup format, which represents the signal map file for the whole genome in which alignment results (coverage depth) are represented in per-base format. Regions where coverage depth was greater than the lower tenth percentile of expressed genes were considered significantly expressed [21] ; in the current study, this corresponded to a coverage depth of 7 reads/bp in pileup format. As another measure for estimating background expression level, we analyzed the coverage in the intergenic regions of the genome. We assumed that at least half of the intergenic region is not expressed (considering the presence of known expressed regions, such as 39 and 59 UTR of genes, intergenic region of the operons, and sRNAs) and calculated the coverage, which corresponded to #6 reads per base, lower than our first cutoff estimate. We retained the most conservative cutoff for expression, i.e., 7 reads per base for describing the expression map of H. somni. Nucleotides in the genome sequence with coverage depth above our threshold value were considered to be expressed. This resulted in the generation of a whole genome transcriptome profile of H. somni 2336 at a single nucleotide resolution. Figure 1 show the steps involved in the analysis of expressed intergenic regions. We compared the RNA-Seq based transcriptome map with the available genome annotation to identify expressed, novel, and intergenic regions in the genome. Promoters and terminators were predicted across the genome to add confidence to the identified novel elements. For the first time, we report the identification of 94 sRNAs (Table 1) in the H. somni genome. The start and end for sRNA in Table 1 refer to the boundaries of transcriptionally active regions (TAR, putative sRNAs). Of these, twelve were similar to wellcharacterized sRNA families that are described in many bacterial species, such as tmRNA, 6S, and FMN ( Figure 2 ). The total of 82 novel sRNAs reported in this study has not been reported earlier. The majority of the identified sRNAs (.75%) were shorter than 200 nucleotides (length range 70-695 nucleotides). The average GC content of sRNA at 39.3% was slightly higher compared with the 37.4% GC content of the genome. Promoters within 50 nt upstream/downstream of the TAR boundaries were predicted for 68 sRNA. Similarly, Rho-independent transcription terminators were predicted within 50 bp upstream/downstream of 40 sRNA. Figure 3 shows the depth of coverage for one of the identified novel sRNA ''HS46'' viewed in the Artemis genome browser [22] . BLAST analysis of the sRNA sequences against the nonredundant, nucleotide database at NCBI revealed that 31 of the sRNA sequences were unique to the H. somni 2336 genome. Another 41 were highly conserved (.95% identity with .95% coverage) only in H. somni strain 129PT, which is a commensal, preputial isolate. A set of 11 sRNAs were conserved in the related Pasteurellaceae family, which includes genomes such as P. multocida, H. influenzae, H. parainfluenzae, and H. ovis. Only 11 sRNAs were conserved in distant bacterial genomes from genera Streptococcus, Clostrodium, Actinobacillus, Vibrio, and others. This lack of sRNA sequence conservation beyond the species could indicate that sRNA sequences are under strong selection pressure, and that they could be responsible for the adaptation of many species to different environmental niches. We searched all H. somni sRNA sequences against the Rfam database [23] to determine their putative functions. We found that 12 sRNAs were homologs to well characterized sRNAs in other genomes. The identified functional categories included FMN riboswitches, gcvB, glycine, intron_gpII, lysine, alpha_RBS, LR-PK1, isrK, MOCORNA, RNaseP_bact_a, tmRNA, and 6S. sRNAs for which no Rfam function could be predicted represent a completely novel set of non-coding sRNAs. Functions of these novel sRNA need to be determined by further experiments. We evaluated the coding potential of all expressed intergenic regions, by conducting BLASTX based sequence searches against the non-redundant protein database at NCBI followed by manual analysis and interpretation. We identified 38 novel protein coding regions ( Table 2 ). The average length of the identified novel proteins was around 60 amino acids (ranged from 19 to 135 amino acids). The majority of the novel proteins (30) were conserved hypothetical proteins present in related species such as H. somni 129PT, M. haemolytica, and H. influenzae. Some of the novel proteins had predicted functions, such as DnaK suppressor protein, toxic membrane protein TnaC, and predicted toxic peptide ibsB3 ( Table 2 ). Figure 4 shows an example of a novel protein ''HSP7'' that is similar (74% similarity and 100% coverage) to a putative, phage-related DNA-binding protein of Neisseria polysaccharea. The single nucleotide resolution map described in this study enabled us to correct the start site for five genes based on the current genome annotation (Table 3) . These genes were annotated as phospholipid synthesis protein, ribosomal protein S2, aconitate hydratase 2, peptide chain release factor 2, and DUF411, a protein of unknown function. Based on evidence from RNA-Seq data, we performed a BLAST comparison with other phylogenetically similar proteins to confirm the new gene boundaries (Table 3) . The comparison of the transcriptome map of the H. somni genome with predicted proteins revealed the presence of frameshift mutations. Four genes have non-functional start codons, resulting in a predicted protein, truncated at the amino terminus (based on BLAST comparison with homologous proteins in other species), although full length mRNA was present. An example is presented for the gene ''HSM_0748'', annotated as ''Alpha-Lfucosidase'' ( Figure S1 ). The other three genes, HSM_0603, HSM_1666 and HSM_1668, encode a hypothetical protein, type III restriction protein res subunit, and CTP synthase, respectively. Two genes with frameshifts causing protein truncations (based on BLAST comparison with homologous proteins) are HSM_1385 (beta-hydroxyacyl dehydratase, FabA) and HSM_1744 (alcohol dehydrogenase zinc-binding domain protein). The transcriptome map revealed a full length mRNA for these two genes that code for truncated proteins. Our transcriptome map of H. somni identified expression from 1636 (approximately 80%) of the predicted genes. The expressed genes were distributed evenly across all TIGRFAM functional categories (Table S1 ). The transcriptome map allowed identification of operon structures at a genome scale, critical for identifying co-expressed genes and for understanding coordinated regulation of the bacterial transcriptome. We identified co-expression for 452 pairs (total 730 genes) of H. somni genes ( Table S2 ) that were transcribed together and constituted a minimal operon. By joining consecutive overlapping pairs of co-expressed genes, we identified 278 distinct transcription units (Table S3) . We compared our experimentally identified co-expressed genes with computationally predicted operons. The overlap between computational prediction of co-expressed genes using DOOR [24] and this study was 86% (394 gene pairs) (Table S4) . Thus, our dataset validates expression of 394 computational gene-pair predictions. We identified 59 new gene pairs that are co-expressed and were not predicted by DOOR, which could be part of unidentified, new operon structures. For example, further in-depth analysis indicated a new operon consisting of three genes: HSM1354, HSM1355 and HSM1356, annotated as ribosomal protein L20, ribosomal protein L35, and translation initiation factor IF-3 respectively, which were not predicted computationally ( Figure 5 ). The orthologs of these genes are well known to form a functional operon of ribosomal proteins (IF3-L35-L20) in Escherichia coli [25] . In this study using RNA-Seq we describe the whole genome transcriptome profile of H. somni 2336, a bovine respiratory disease pathogen. The single nucleotide resolution map helped uncover the structure and complexity of this pathogen's transcriptome and led to the identification of novel, small RNAs and protein coding genes as well as gene co-expression. Prokaryotic genome annotation is performed often using computational gene prediction programs [8, 9] . However, these prediction algorithms are not able to identify the non-coding sRNAs, antisense transcripts, and other small proteins. To overcome the shortcomings of computational genome structural annotation, various experimental methods are used for identification of novel expressed elements [13, 14, 15, 16, 17, 18, 26, 27, 28] . Deep transcriptome sequencing (RNA-Seq) has emerged recently as a method that enables the study of RNA-based structural and regulatory regions at the genome scale. RNA-Seq technology has many advantages compared with existing array based methods for transcriptome analysis. In particular, RNA-Seq does not require probes, so the process is free from probe design issues or bias from hybridization issues. Also, the transcriptome coverage from RNA-Seq is very high [29, 30] . RNA-Seq was demonstrated to be effective for the discovery of bacterial non-coding RNAs, accurate operon definition, and correction of gene annotation [27, 31, 32] . Therefore, in the current study, we used RNA-Seq for profiling H. somni 2336 transcriptome. Mapping of RNA-Seq reads onto the H. somni genome sequence resulted in more than 94% coverage with at least one read per base. This observation is consistent with the reported 94% genome expression in Bacillus anthracis, 89.5% in Sulfolobus solfataricus, and 95% in Burkholderia cenocepacia, studied under one or more experimental growth conditions using RNA-Seq [32, 33, 34] . These results indicate that most of the bacterial genome sequence is expressed at some basal level. To identify significantly expressed regions above this baseline, we used two alternative methods (discussed in Results section) to estimate the background expression. Both methods yielded similar results (6-7 reads per base). We selected the higher stringency cutoff of 7 reads per base to minimize the number of false positives. We identified a total of 95 sRNAs in the H. somni genome. Twelve of these were predicted by Rfam [23] and are similar to conserved sRNA (e.g., 6S, tmRNA, FMN) in other bacterial species, which helps validate our approach. The 83 novel H. somni sRNAs may have housekeeping function, regulatory activity, or participate in virulence as described in other pathogenic bacteria [19, 35, 36] . The identified sRNAs did not show any location specific bias across the genome. Similarly, genes known to be associated with virulence are known to be scattered across bacterial genomes [37, 38] . However, the tendency to form clusters was observed with sRNAs, which could indicate that functionally related sRNAs tend to be located in close proximity. The RNA-Seq based transcriptome map of H. somni identified 38 novel protein coding genes that were missed by the initial annotation. The average length of the proteins coded by these genes exceeds 60 amino acids, suggesting that length based cutoff was not the main reason that these genes were missed by computational gene prediction programs. The novel protein coding genes identified in the current study could serve as a training set to improve gene prediction algorithms. The transcriptome map helped to identify incorrect annotation of start codons in the genome. Transcriptional mapping does not provide direct evidence of translational start sites. However, location of identified transcriptional start sites suggest that the annotated start codons are incorrect, an observation that is confirmed by BLAST comparisons against homologous genes in other bacterial species. Transcriptional mapping revealed genes where the 59 untranslated sequence extended well beyond the translational start. BLAST comparisons indicated that these genes have either nonsense or missense base changes relative to homologous genes in other bacterial species, causing apparent ''truncated'' proteins compared with those in other species. Further work is needed to determine whether these 59 untranslated regions serve regulatory functions or they are vestigial. RNA-Seq data enabled us to determine operon structures at a genome scale, and it allowed identification of some operons not predicted by the computational operon prediction method. Operon structures that include genes not expressed under the experimental growth condition used in the current study, could not be identified. Our results support the notion that using a combination of experimental operon identification by RNA-Seq and computational prediction can improve operon identification in bacterial genomes [39] . For the first time, we report the RNA-Seq based transcriptome map of H. somni 2336 and describe novel expressed regions in the genome. Whereas the results are interesting, we are aware of the limitations of the study. Because the RNA-Seq protocol was not strand specific, we could not determine the strand specificity of expressed novel transcripts. Therefore, Table 1 lacks information about sRNA orientation in the genome. Because strand specific information was missing, we could not describe antisense expression in the genome. For protein coding genes, we derived strand specificity based on alignment of the BLAST hit. Despite this shortcoming, we identified novel expressed regions and transcriptional patterns across the whole genome at a high coverage, which is not possible by other transcriptome analysis methods. Overall, this study describes RNA-Seq based transcriptome map of H. somni for identification of functional elements in a pathogen of importance to agriculture. Our genome-wide survey predicts numerous, novel, expressed regions that need biological characterization for understanding disease pathogenesis. Description of all functional elements in the H. somni system is a prerequisite for conducting holistic systems approaches to understand the complex pathogenesis of bovine respiratory disease. We propagated H. somni 2336 on three TSA-blood plates (with 5% sheep red blood cells) for 16 hr or until a fresh lawn of cells was visible. IBC approval was not required for acquiring the plates as they were purchased through a commercial vendor: Fisher Scientific (Pittsburgh, PA), and manufactured by Becton Dickinson Diagnostic Systems, (Franklin Lakes, NJ). We washed the plates with brain heart infusion (BHI) broth, adjusted the culture to an OD620 nm = 0.8, and supplemented with RNAprotect reagent. The cells were harvested by centrifugation and stored at 280uC. We extracted total RNA using the RNeasy mini kit (Qiagen, Valencia, CA) following the manufacturer's protocol. Total RNA was treated with RNase-free DNAse (Invitrogen, Carlsbad, CA). Using Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA), we determined the RNA integrity number (RIN) of total RNA to be greater than 8. MICROBExpress TM Kit (Ambion, TX, USA), which specifically removes rRNAs, was used for mRNA enrichment. Small RNAs (i.e., tRNA and 5S rRNA) are not removed with this enrichment step (confirmed by Bioanalyzer). We used 100 ng enriched mRNA with Illumina mRNA-Seq sample preparation kit (Illumina, San Diego, CA) for library construction following the manufacturer's protocols. Briefly, mRNA was fragmented chemically by divalent zinc cations and randomly primed for cDNA synthesis. After ligating paired-end sequence adaptors to cDNA, we isolated fragments of approximately 200 bp by gel electrophoresis and amplified. We We checked all Illumina reads for quality, and removed sequence reads containing ''Ns''. Custom perl script was written to convert Illumina reads into fastq format. The script ''fq_all2std.pl'' from MAQ [40] converted fastq format to Sanger fastq format. Reads in sanger fastq format, were mapped onto the Histophilus somni 2336 genome sequence (GenBank Accession number. CP000947) using the alignment tool Bowtie [41] , allowing for a maximum of two mismatches. The reads that mapped to more than one location were discarded. We used Samtools [42] to convert data into SAM/BAM format, and to generate alignment results in a pileup format. Pileup format provides the signal map file and has per-base format coverage. Custom perl scripts were written to calculate the background expression. Processed data was deposited in GEO with the accession number GSE29578. We used in-house perl scripts to extract novel expressed intergenic regions to identify novel small RNAs, riboswitches, and putative novel proteins. sRNA ,70 bp in length were discarded to minimize the number of false positives. For each novel expressed region, BLAST sequence searches were performed against the non-redundant protein database at NCBI to identify potential protein coding regions. Intergenic regions within predicted operons [24] represent expressed regions and can be mis-classified as sRNAs. Therefore, these regions were excluded. We analyzed BLAST results manually, to identify novel protein coding regions and start codon corrections. If no protein coding region was found in the intergenic expressed regions, the presence of a promoter or a rho-independent terminator allowed us to classify the regions as sRNA. Bacterial promoter sequences were predicted by Neural Network Promoter Prediction program (http://www.fruitfly.org/seq_tools/promoter.html) [43] . Rho-independent transcription terminators were identified using the program TransTermHP [44] . For functional annotation, all identified identified sRNA sequences were searched against the Rfam database [23] . sRNA sequence conservation among other genomes was determined by blastn searches against nonredundant nucleotide database at NCBI. We mapped sRNAs, along with additional features, onto genome browsers like IGV [45] and Artemis [46] for further visualization, manual analysis, and interpretation. Gene expression: expressed reads with coverage above background were mapped onto the annotated genes of H. somni 2336. Genes that had a significantly higher proportion of their length (.60%) covered by expressed reads were considered to be expressed. Operons: RNA-Seq can identify and predict operon structures in bacteria. We considered two or more consecutive genes to be part of an operon, if they fulfilled the following criteria: (a) they are expressed; (b) they are transcribed in the same direction; and (c) the intergenic region between the genes is expressed. Overlapping pairs of such genes were joined together to identify large operon structures. We used in-house perl scripts for the analyses. Figure S1 Mutated start codon. The Figure shows that the predicted protein coding frame (MH_748) is shorter at the 59 end than the corresponding transcript level shown by the RNA-Seq coverage. Although the transcript is longer near 59 end, no start codon is found in that region which might be a result of the mutation in that region of the start codon. This was further validated using homology searches of the full length transcript which shows high homology (95% Identity and .95% coverage) to a alpha-L-fucosidase protein from M. haemolytica PHL213. (TIF)

Molecular mechanism for 3:1 subunit stoichiometry of rod cyclic nucleotide-gated ion channels

Molecular determinants of ion channel tetramerization are well characterized, but those involved in heteromeric channel assembly are less clearly understood. The heteromeric composition of native channels is often precisely controlled. Cyclic nucleotide-gated (CNG) channels from rod photoreceptors exhibit a 3:1 stoichiometry of CNGA1 and CNGB1 subunits that tunes the channels for their specialized role in phototransduction. Here we show, using electrophysiology, fluorescence, biochemistry, and X-ray crystallography, that the mechanism for this controlled assembly is the formation of a parallel 3-helix coiled-coil domain of the carboxy-terminal leucine zipper region of CNGA1 subunits, constraining the channel to contain three CNGA1 subunits, followed by preferential incorporation of a single CNGB1 subunit. Deletion of the carboxy-terminal leucine zipper domain relaxed the constraint and permitted multiple CNGB1 subunits in the channel. The X-ray crystal structures of the parallel 3-helix coiled-coil domains of CNGA1 and CNGA3 subunits were similar, suggesting that a similar mechanism controls the stoichiometry of cone CNG channels. SUPPLEMENTARY INFORMATION: The online version of this article (doi:10.1038/ncomms1466) contains supplementary material, which is available to authorized users.

T he precise assembly of many ion channels relies on a multimerization domain that assembles compatible subunits and excludes incompatible subunits. In the voltage-gated family of channels, these multimerization domains take the form of either an amino-terminal tetramerization domain (T1 domain) 1,2 or a carboxy-terminal four helix coiled-coil domain [3] [4] [5] . Although channel assembly is generally permitted without these multimerization domains, such domains confer subunit-specific assembly, often restricting the channel composition to like-subunits. Cyclic nucleotide-gated (CNG) channels, however, have a more complex subunit composition. Native CNG channels are composed of a stereotyped number of CNGA subunits and CNGB subunits. CNGA and CNGB subunits are structurally related and both are thought to contribute to the formation of the pore. The CNG channel from rod photoreceptors is composed of three CNGA1 subunits and one CNGB1 subunit [6] [7] [8] , whereas the CNG channel from olfactory receptors is composed of two CNGA2 subunits, one CNGA4 subunit and one CNGB1b subunit 9 . The cone CNG channels comprise CNGA3 subunits and CNGB3 subunits, though the subunit stoichiometry is still unknown 8, 10 . The presence of the CNGB subunits confers enhanced Ca 2 + permeation, modulation by Ca 2 +calmodulin, and native-like cyclic-nucleotide specificity compared with homomeric CNGA channels [11] [12] [13] [14] [15] [16] [17] . The precisely controlled subunit compositions, therefore, tune the CNG channels for their specialized roles in phototransduction and olfactory transduction. CNG channels are activated by the direct binding of intracellular cyclic nucleotides 18 . They are members of the voltage-gated ion channel superfamily, consisting of four subunits around a centrally located pore. The intracellular carboxy-terminal region of each subunit consists of a C-linker region, a cyclic nucleotide-binding domain (CNBD), and a post-CNBD region (Fig. 1a) . The Xray crystal structure of the carboxy-terminal region of the related hyperpolarization-activated cyclic nucleotide-gated channel reveals a four fold symmetric tetramer with non-interacting CNBDs and extensive intersubunit interactions between C-linker regions 19 . How does the complex subunit composition of CNG channels come about? The CNGA subunits contain a carboxy-terminal leucine zipper (CLZ) domain in the post-CNBD region (Fig. 1a,b) 8 . On the basis of the finding that the CLZ domain from CNGA3 forms homotypic trimers in solution, it was proposed that trimerization of this domain constrains the channel to contain three CNGA subunits 8 . Previously, it has been shown that point mutations in this domain decreased the proportion of heteromeric channels (relative to CNGA homomers) but did not alter the 3:1 stoichiometry of the heteromeric channels 20 . Here we test the role of the CLZ domain by measuring the subunit composition of CNGA1 + CNGB1 channels when the CLZ domain in the CNGA1 subunit was deleted. Using electrophysiology and fluorescence measurements, we found that deletion of the CLZ domain relaxed the constrained subunit stoichiometry. This allowed more than one CNGB1 subunit in the channel and conferred novel pharmacological properties and cyclic nucleotide specificity. Using X-ray crystallography, we found that the CLZ domains of both CNGA1 and CNGA3 formed homotypic parallel 3-helix coiled-coil domains, consistent with their proposed role in regulating subunit assembly. To determine the role of the CLZ domain in controlling heteromeric assembly of CNGA1 + CNGB1 channels, we first measured the functional effects of deleting the CLZ domain. CNG channels were expressed in Xenopus oocytes and studied using the inside-out configuration of the patch-clamp technique. Homomeric channels of CNGA1 subunits are efficiently activated by saturating cGMP, but poorly activated by saturating cAMP (I cAMP /I cGMP = 0.05 ± 0.01, n = 6) (Fig. 2a , black is with 2.5 mM cGMP, red is with 16 mM cAMP). However, incorporation of a single CNGB1 subunit per tetramer confers a significant elevation in cAMP efficacy (I cAMP / I cGMP = 0.22 ± 0.03, n = 7) (Fig. 2a) . The mechanism of the increased cAMP efficacy by the CNGB1 subunit is unknown but could involve either cAMP binding to CNGB1 or simply a more favourable opening transition with CNGB1 (or both). This increased efficacy is constant over a wide range of CNGB1 expression levels, indicating that the CNGB1 subunit incorporates into the channel with a fixed and tightly-regulated stoichiometry 15 . Previously, it has been shown that the amino-terminal region of the CNGB1 subunit interacts with the CLZ domain of the CNGA1 subunit and regulates trafficking of heteromeric channels 21 . To test for a possible role of this interaction in subunit assembly, we deleted the amino-terminal region (amino acids 2-765) of CNGB1 (CNGB1-∆N). Co-expressing CNGA1 and CNGB1-∆N produced robust currents with amplitude similar to that of CNGA1 + CNGB1 (ref. 21 ). In addition, as shown in Figure 2a , deletion of the aminoterminal region of CNGB1 had no effect on the cAMP efficacy of the heteromeric channels (I cAMP /I cGMP = 0.21 ± 0.03, n = 7). These results suggest that the amino-terminal domain of CNGB1 is not required for the proper assembly of functional heteromeric channels. Next, we tested the effects of deletion of the CLZ domain (amino acids 609-693) of the CNGA1 subunit (CNGA1-∆CLZ). Deletion of the CLZ domain had no effect on the expression or measured functional properties of CNGA1-∆CLZ homomeric channels (Fig. 2a) 21 . However, it had a dramatic effect on expression and functional properties of CNGA1-∆CLZ + CNGB1 heteromeric channels. As shown previously, co-expression of CNGA1-∆CLZ with CNGB1 at a typical 1:4 RNA ratio virtually eliminated all functional channel expression (Fig. 2a) 21 . This dominant negative effect may underlie one form of retinitis pigmentosa resulting from deletion of a portion of the CLZ domain in CNGA1 (refs 21,22) and suggests that the CNGA1-∆CLZ channels are still interacting with CNGB1, perhaps in an uncontrolled manner. To recover functional expression of heteromeric channels, we co-injected CNGA1-∆CLZ with CNGB1-∆N containing a deletion of the trafficking signal in the amino-terminal region of CNGB1 (ref. 21) . Even at an RNA ratio of 1:2 CNGA1-∆CLZ: CNGB1-∆N, the effect of the CLZ domain deletion on the heteromeric channels was striking. Co-expression of CNGA1-∆CLZ and CNGB1-∆N yielded channels with a highly elevated cAMP efficacy (I cAMP /I cGMP = 0.38 ± 0.02, n = 10) (Fig. 2a) , significantly higher than CNGA1 homomeric channels and CNGA1 + CNGB1 heteromeric channels (Fig. 2b) . The elevated cAMP efficacy seen with the CLZ domain deletion in heteromeric channels, but not homomeric channels, suggests the possibility that CNGA1-∆CLZ and CNGB1-∆N form heteromeric channels with a novel stoichiometry, perhaps with more than one CNGB1 subunit per channel. To further test for incorporation of multiple CNGB1 subunits on deletion of the CLZ domain, we measured the pharmacological properties of heteromeric channels. Although 100 µM L-cisdiltiazem is ineffective at blocking CNGA1 homomers, it potently blocks CNGA1 + CNGB1 heteromers (Fig. 3a , black is control, blue is 100 µM L-cis-diltiazem) 17 . The block is voltage dependent with a Hill slope of one, suggesting a pore-blocking mechanism (Fig. 3a,b) . The CNGA1-∆CLZ mutation had no effect on block of homomeric channels, and the CNGB1-∆N mutation had no effect on block of heteromeric channels. Surprisingly, the block of CNGA1-∆CLZ + CNGB1-∆N heteromers was intermediate between CNGA1 (or CNGA1-∆CLZ) homomers and , and CnGA1-∆CLZ + CnGB1-∆n (n = 10). The expected subunit stoichiometry for each case is shown at the bottom. Data are plotted as mean ± s.e.m. Asterisks denote statistically significant differences with P < 0.01. CNGA1 + CNGB1 (or CNGA1 + CNGB1-∆N) heteromers (Fig. 3a) . Previously, it was shown that heteromers formed with point mutations in the CLZ domain of CNGA3 exhibited intermediate block by L-cis-diltiazem because they formed mixtures of CNGA3 homomers and CNGA3 + CNGB1 heteromers 20 . However, that was not the case here. As shown above, CNGA1-∆CLZ + CNGB1-∆N heteromers actually exhibit greater cAMP efficacy than either CNGA1 homomers or CNGA1 + CNGB1 heteromers (Fig. 2) . Together, these results indicate the formation of a previously uncharacterized population of heteromeric CNG channel complexes. To unravel the cause of this intermediate block, we measured the dose-response relations for L-cis-diltiazem block at + 60 mV. The dose-response relations for channels containing CLZ domains were well fit with a single Langmuir isotherm, suggesting a single population of channels (Fig. 3b) . The block of CNGA1 and CNGA1-∆CLZ homomers were similar and low affinity (K 1/2 = 174 ± 7.2 µM, n = 5 and K 1/2 = 164 ± 8.2 µM, n = 6, respectively), and the block of CNGA1 + CNGB1 and CNGA1 + CNGB1-∆N heteromers were similar and high affinity (K 1/2 = 2.45 ± 0.28 µM, n = 5 and K 1/2 = 1.40 ± 0.21 µM, n = 5, respectively) (Fig. 3b) . The dose-response relation for block of CNGA1-∆CLZ + CNGB1-∆N heteromers, however, was not fit well by either a single Langmuir isotherm, nor by the sum of two Langmuir isotherms with affinities constrained to those of CNGA1 homomers and CNGA1 + CNGB1 heteromers. Instead, fits to the sum of two Langmuir isotherms revealed a small high affinity component similar to that of CNGA1 + CNGB1 channels (K 1/2 = 0.56 ± 0.09 µM, n = 4), and a larger component with an affinity intermediate between that of CNGA1 homomers and CNGA1 + CNGB1 heteromers (K 1/2 = 50 ± 5 µM, n = 4) (Fig. 3b) . These results suggest that the heteromers formed from CNGA1-∆CLZ and CNGB1-∆N are not a mixture of CNGA1 homomers and normal CNGA1 + CNGB1 heteromers, but, instead, a mixture of normal 3:1 CNGA1 + CNGB1 heteromers and novel CNGA1 + CNGB1 heteromers. One intriguing possibility is that the novel heteromers formed without a CLZ domain contain multiple CNGB1 subunits. This would also suggest that whereas one CNGB1 subunit confers much greater L-cis-diltiazem affinity, extra CNGB1 subunits lower the L-cis-diltiazem affinity (Fig. 3d) . This model predicts that lowering the relative amount of expressed CNGB1 subunits should actually increase the fraction of the high affinity component (due presumably to channels with the normal 3:1 CNGA1:CNGB1 stoichiometry), because a lower amount of CNGB1 should favour a subunit stoichiometry with fewer CNGB1 subunits. The results of such an experiment are shown in Figure 3c . Lowering the RNA co-injection ratio of CNGA1-∆CLZ and CNGB1-∆N from 1:2 to 1:1 significantly increased the fraction of the high affinity component from 0.21 ± 0.01, n = 3 to 0.40 ± 0.01, n = 4. In addition, the dose-response relation is now poorly fit by a sum of two Langmuir isotherms and is suggestive of a third component with an affinity similar to CNGA1 homomers. Together, these results suggest that deletion of the CLZ domain removes the constraint allowing only one CNGB1 subunit in the channel tetramer. This allows multiple CNGB1 subunits that confer higher efficacy for cAMP and (surprisingly) lower affinity for L-cis-diltiazem block (Fig. 3d ). If channels containing three CNGB1 subunits were not functional or not trafficked properly to the plasma membrane, then this loss of controlled assembly could also explain the large dominant negative effect of CNGB1 subunits, when the CLZ domain is deleted from the CNGA1 subunit. To directly measure the presence of multiple CNGB1 subunits in the channel, we turned to an approach based on fluorescence resonance energy transfer (FRET) 23 . FRET reports the proximity of two fluorophores based on the efficiency of energy transfer between a donor fluorophore and an acceptor fluorophore. FRET can be detected by an enhanced emission of the acceptor fluorescence on excitation of the donor. Cyan fluorescent protein (eCFP) and yellow fluorescent protein (eYFP) are genetically-encoded donor and acceptor fluorophores, respectively, that can be fused to proteins of interest 24 . They exhibit 50% energy transfer at a distance of about 50 Å and measurable transfer up to 80 Å (ref. 25) , making them ideally suited for measurements of subunit composition 7 . Previously it was shown that if CNGA1 subunits are fused to eCFP and eYFP (individually) and co-expressed with CNGB1 subunits in Xenopus oocytes, there was significant FRET for channels in the plasma membrane 7 . However, if CNGB1 subunits were fused to eCFP and eYFP and co-expressed with CNGA1 subunits, there was little or no FRET (Fig. 4a) . These results revealed that heteromeric rod CNG channels in the plasma membrane contain multiple CNGA1 subunits but only a single CNGB1 subunit (3:1 stoichiometry). We performed a similar experiment to determine if deletion of the CLZ domain permitted incorporation of multiple CNGB1 subunits into the heteromeric channels. If there are multiple CNGB1 subunits in the heteromeric channels, then we should observe FRET between CNGB1 subunits when fused to eCFP and eYFP and co-expressed with CNGA1-∆CLZ subunits (Fig. 4a) . For these experiments, we fused eCFP and eYFP (individually) to the amino-terminal end of CNGB1-∆N (eCFP-CNGB1-∆N and eYFP-CNGB1-∆N respectively). The eCFP and eYFP fusion constructs exhibited functional properties indistinguishable from their non-fusion counterparts (data not shown). As the amino-terminal region is deleted in this subunit, the fluorescent proteins should be near the inner leaflet of the membrane, close to the S1 transmembrane segment. Based on the structure of the related Kv1.2 channel 26 , two fluorescent proteins in the same channel should be about 56 Å apart (assuming they are in adjacent subunits) or 79 Å apart (assuming they are in nonadjacent subunits). Representative results from co-expression of CNGA1-∆CLZ with eCFP-CNGB1-∆N and eYFP-CNGB1-∆N are shown in Figure 4b . FRET was measured from the sensitized emission of the acceptor using fluorescence spectra 23 . The fluorescence emission spectra from channels in or near the plasma membrane of live Xenopus oocytes were measured with confocal microscopy. Excitation of eCFP with a 458 nm laser produced a complex emission spectrum (red trace) with both eCFP and eYFP emission components. The eYFP component (orange trace) was extracted by subtracting off a scaled emission spectrum from eCFP alone (blue trace). This extracted spectrum contains eYFP emission from both FRET and direct excitation of eYFP by 458 nm light. To isolate the component due to FRET, the fractional excitation of eYFP by 458 nm light relative to 488 nm light (Ratio Ao) (Fig. 4c, Methods) was subtracted off the corresponding fractional excitation for eCFP and eYFP-containing channels (Ratio A). Ratio A-Ratio Ao is directly proportional to the FRET efficiency 23 . As shown in Figure 4d , Ratio A-Ratio Ao was significantly larger with CNGA1-∆CLZ than with CNGA1, when the CNGA1 subunit was coexpressed with fluorescent CNGB1. These results directly demonstrate that deletion of the CLZ domain caused multiple CNGB1 subunits to assemble. A possible alternative explanation for the FRET seen between CNGB1 subunits with deletion of the CNGA1 CLZ domain is that the CNGB1 subunits are forming homomeric channels at low levels. Indeed, the fluorescence is considerably dimmer with co-expression of CNGA1-∆CLZ than with CNGA1. Previously, however, it has been shown that, in the absence of CNGA1, CNGB1 subunits do not form functional homomeric channels and are only very weakly expressed in the plasma membrane 9, 17 . Similarly, we found that, in the absence of a CNGA1 subunit, both CNGB1 and CNGB1-∆N did not produce detectable currents and are poorly expressed in the plasma membrane ( Supplementary Fig. S1 ). In addition, it was shown previously that the CNGB1 subunits expressed alone do not exhibit appreciable FRET 9 . This makes it unlikely that the increased FRET seen between CNGB1 subunits with deletion of the CNGA1 CLZ domain is from CNGB1 subunits forming homomeric channels. Nevertheless, to prove that multiple CNGB1 subunits are coassembled with CNGA1-∆CLZ subunits, we examined the FRET between CNGA and CNGB subunits. For acceptor sensitization, the apparent FRET efficiency increases as the donor-to-acceptor ratio increases 27 . Therefore, if the donor fluorophore (eCFP) was fused to the CNGB subunit and the acceptor fluorophore (eYFP) was fused to the CNGA subunit, then an increase in apparent FRET should be seen if multiple CNGB1 subunits assembled in the channel complex ( Supplementary Fig. S2) . Indeed, the results show a significant increase in FRET between eCFP-CNGB1-∆N and eYFP-CNGA1-∆CLZ compared with eYFP-CNGA1 (Fig. 4e) . These results indicate that multiple CNGB1 subunits are co-assembling with CNGA1-∆CLZ subunits. Overall, it is clear that deletion of the CLZ domain relaxes the 3:1 constraint on subunit assembly. The above results suggest that the CLZ domain has an important role in constraining the stoichiometry of rod CNG channels to 3:1 CNGA1:CNGB1 subunits. Previously, it has been shown that a protein fragment containing the CLZ domain from cone CNGA3 forms homotypic trimers in solution 8 . To determine whether the CLZ domain from CNGA1 would also form trimers, and to localize the region involved, we screened a number of carboxy-terminal fragments of both CNGA1 and CNGA3 using fluorescence-detection size-exclusion chromatography (FSEC) (Supplementary Fig. S3) 28 . The protein fragments were expressed in bacteria as N-terminal (NGFP) or C-terminal (CGFP) fusions to GFP. After cell lysis, the crude supernatant was then loaded on a size-exclusion chromatography system with an inline fluorescence detector. All the expressing fragments that contained the CLZ domain showed an appreciable component that eluted at 13 mL, consistent with the formation of a trimer (Fig. 5a) . To verify the composition of the high molecular weight component, samples of the purified protein, following cleavage of GFP, were analysed by light scattering size-exclusion chromatography (LS-SEC) (Fig. 5b) 29, 30 . For both CNGA1 and CNGA3, the molecular weights determined from LS-SEC were within about 1% of the predicted molecular weights for a trimer (Table 1) . Similar results with smaller fragments revealed a minimal fragment, corresponding to the CLZ domain, that formed trimers in both CNGA1 (amino acids 621-667) and CNGA3 (626-672) subunits. The sequence of the CLZ domain exhibits periodic heptad repeats (a-b-c-d-e-f-g) n in which the 'a' and 'd' positions are hydrophobic residues (Fig. 1b) 8 . This motif is characteristic of coiled-coil domains 31, 32 . There are two noncontiguous heptad repeat regions in each sequence; the first has three repeats and the second has two repeats. The two regions are out of register, with the 'a' position of the first region becoming the 'd' position of the second region. While the presence of heptad repeats is fairly indicative of coiledcoil domains, it is difficult to determine from the sequence alone the oligomeric state, parallel versus antiparallel arrangement, and details of the coiled-coil packing and surface architecture 32 . These properties have direct bearing on the mechanism of assembly, the interpretation of mutations, and the interaction with other proteins or domains. For these reasons, we were interested in obtaining the high-resolution structure of the CLZ domains. We have solved the high-resolution X-ray crystal structures of the CLZ domains of both CNGA1 and CNGA3. The proteins were expressed and purified from bacterial cell lysates and crystallized under vapour diffusion. Crystals of CNGA1#621-690 grew in space group P2 1 2 1 2 1 with three molecules in the asymmetric unit and diffracted to 2.14 Å. The structure of CNGA1#621-690 was solved using molecular replacement with the structure of HIV gp41 N-trimer pocket region 33 (3L36) as a search probe (Supplementary Table S1 ). The structure of CNGA3#626-672 in space group P12 1 1 was solved to 1.9 Å using molecular replacement with the structure of SARS virus S2 protein 34 (1ZVB) as a search probe (Supplementary Table S1 ). The structures of both CNGA1#621-690 and CNGA3#626-672 consisted of long (65 Å) parallel three-helix coiled-coil domains (Fig. 6) . The sequence beyond the CLZ domain in CNGA1 (amino acids 665-675 for chain A; 665-676 for chain B; and 665-670 for chain C) flared outward and was partially involved in crystal packing interactions (Supplementary Fig. S4 ). In addition, the CNGA1#621-690 structure contained two bound heavy atoms per trimer with an electron density greater than 10 sigma ( Fig. 6 ; Supplementary Fig. S5 ). These heavy atoms were coordinated by acidic residues (E629 and D633 for the proximal site, and E671 and E623 for the distal ion) at the interface between different trimers in the crystal and, therefore, are presumed to be non-physiological ( Supplementary Fig. S5 ). They are likely to be Zn 2 + ions because of their coordination and the requirement for Zn 2 + in the crystallization solution to produce crystals. The packing of the coiled-coil domains followed the characteristic 'knobs into holes' arrangement with the 'a' and 'd' positions forming alternating layers of the hydrophobic core of the coiled coil ( Fig. 7) 32, 35 . The two noncontiguous heptad repeat regions assembled into a single continuous coiled coil. Interestingly, the CNGA1 and CNGA3 structures differed in the length of the two heptad repeat regions. For CNGA1, the first region contained 3 heptad repeats and the second contained 3.5 repeats (Fig. 7b) . For CNGA3 the first region contained 4 heptad repeats and the second contained 2.5 repeats (Fig. 7c) . The difference appears to be the substitution of an isoleucine for a leucine after the third heptad repeat in CNGA1 (Fig. 1b) . Although the structures of CNGA1 and CNGA3 were generally similar, they were not identical. As stated above, different residue positions were used in the helical packing between the two heptad repeat regions in the sequence. In addition, the supercoiling was different between CNGA1 and CNGA3, with a somewhat longer pitch for CNGA1 (181 Å) than for CNGA3 (149 Å) (Supplementary Table S2 ). These differences were also reflected in the structural alignments of the individual subunits of CNGA1 and CNGA3 (rmsd of Cα ranging from 0.87 Å to 2.0 Å). These differences in structure, however, do not have any known functional consequences. The results from the above experiments, and those of others, suggest a molecular mechanism for the controlled assembly of rod CNG channels with a 3:1 stoichiometry of CNGA1:CNGB1 subunits (Fig. 8a) [6] [7] [8] [9] 20 . In this mechanism, the CNGA1 subunits initially assemble into trimers through homotypic association of their carboxy-terminal CLZ domains into a parallel three helix coiled coil. Subsequently, a CNGB1 subunit, if present, is preferentially incorporated into the remaining slot of the channel tetramer because of a high affinity association with CNGA1 subunits. This mechanism incorporates aspects of two previous models for heteromeric assembly of CNG channels: initial assembly of CNGA1 trimers 8 , and high affinity association between CNGA1 and CNGB1 subunits 9,21 . The trimerization of the CLZ domain imposes the constraint that the channels contain at least three CNGA1 subunits, whereas the high affinity association between CNGA1 and CNGB1 assures that the fourth subunit is CNGB1. In support of this mechanism, we showed with electrophysiology and fluorescence that deletion of the CLZ domain in CNGA1 causes multiple CNGB1 subunits to incorporate into the heteromeric CNG channel. Furthermore, with X-ray crystallography we showed that there is a trimerization of the CLZ domain for both CNGA1 and CNGA3 (Fig. 5) that results from the formation of a parallel three helix coiled coil (Fig. 6) . In an elegant series of experiments, Yau and co-workers have proposed a mechanism for 3:1 CNGA:CNGB assembly that involves initial assembly of CNGA trimers 8 . They show that other three helix coiled coils, but not two helix or four helix coiled coils, could substitute for the CLZ domain of CNGA channels to reproduce normal heteromeric channel function and expression 20 . However, whereas we found that deletion of the CLZ domain causes an increase in CNGB1 incorporation in the channels, they showed that mutations in the CLZ domain decreased the incorporation of CNGB1 subunits 20 . Their results were explained by an assembly model where excess CNGA1 monomers outcompete CNGB1 monomers for channel incorporation. Instead, we propose that CNGB1 monomers outcompete CNGA1 monomers for channel incorporation owing to a high affinity association of CNGA1 and CNGB1 subunits. A high affinity association between CNGA and CNGB subunits has been previously proposed to explain the dependence of the CNGA:CNGB ratio in the surface membrane as a function of the ratio of their RNAs injected into the oocyte 9 . A high affinity association of CNGA1 and CNGB1 subunits also explains the large dominant negative effect of CNGB1 on deletion of the CLZ domain in CNGA1 (Fig. 2a) 21 . When the CLZ domain is deleted, the CNGA1 subunits preferentially interact with CNGB1 subunits, resulting in channels with multiple CNGB1 subunits that are not as efficiently expressed in the plasma membrane (Fig. 8b) . This defect has been proposed to underlie one form of inherited retinitis pigmentosa resulting from deletion of the last two heptad repeats of the CLZ domain of CNGA1 (refs 21,22) . Previously, it has been shown that deletion of the aminoterminal region of CNGB1 could partially restore expression of these CNGA1-∆CLZ-containing channels 21 , suggesting that the amino-terminal region of CNGB1 might be involved in the highaffinity association between CNGA1 and CNGB1 subunits. Alternatively, the amino-terminal region of CNGB1 could contain a trafficking signal for incorrectly assembled heteromeric CNG channels 21 . This trafficking signal could provide a mechanism for error-correction where channels with more than one CNGB1 subunit are either poorly trafficked to the plasma membrane or rapidly removed from the membrane. Beyond channel assembly, the CLZ coiled-coil domain may also serve as a scaffold for interactions with other proteins. Previously, it has been shown that the calmodulin binding site in the aminoterminal region of CNGB1 directly interacts with the CLZ domain of CNGA1 (refs 36,37) . Furthermore, Ca 2 + -calmodulin disrupts this interaction resulting in separation of the amino-terminal region of CNGB1 from the carboxy-terminal region of CNGA1 (ref. 38) . Also, the CLZ coiled-coil domain may be an important site of interaction for protein complexes that regulate channel activity, as has been proposed for Kv7 channels [39] [40] [41] . Although the subunit stoichiometry of the rod CNG channel has been well established, the stoichiometry of the cone CNG channel is still uncertain. Whereas some studies have suggested it also has a 3:1 stoichiometry of CNGA3:CNGB3, others have suggested a 2:2 stoichiometry 8, 10 . Here we find that the CLZ domain of CNGA3 forms a parallel three helix coiled coil that is similar to that of CNGA1. This suggests that the stoichiometry and mechanism of assembly for the cone CNG channel will be similar to that of the rod CNG channel. In addition, it has been shown that other three helix coiled-coil domains can substitute for the CLZ domain in cone CNG channels, but two and four helix coiled-coil domains cannot 20 . Interestingly, the olfactory CNG channel has a stoichiometry of 2:1:1 CNGA2:CNGA4:CNGB1b 9 . Both the CNGA2 and CNGA4 subunits have a CLZ domain (Fig. 1b) . It remains to be determined whether the 2:1:1 stoichiometry comes about from a heterotypic three helix coiled coil of the CNGA2 and CNGA4 subunits, or from other intersubunit interactions such as seen at the level of the C-linker regions in the related hyperpolarization-activated cyclic nucleotide-gated channels 19 . In addition, the 3:1 stoichiometry of the TRPP2/PKD1 complex has recently been proposed to involve a trimeric coiled-coil domain in the carboxy-terminal region of TRPP2 (ref. 42) , suggesting that the assembly mechanism of CNG channels may extend to other distantly related proteins. Molecular biology. The bovine CNGA1 complementary DNA used was described previously 43, 44 and contains a C-terminal FLAG epitope (DYKDDYK). The bovine CNGB1 cDNA was a kind gift from Dr. R. Molday 14 and the human CNGA3 cDNA was a kind gift from Dr. K.W. Yau 45 . CNGA1-∆CLZ (CNGA1-∆609-693) and the CNGB1-∆N (CNGB1-∆2-765) were made as described 21 . CNGB1-∆N had a further manipulation, where its internal NcoI restriction site was removed via silent mutation and re-inserted at the initial methionine codon using a PCRbased method. Improved monomeric versions of eCFP and eYFP (mCerulean and mCitrine) were employed in the generation of all fluorescent protein fusions with CNG subunits used for FRET experiments. cDNAs for mCerulean and mCitrine were kindly provided by Dr D Piston 46 and Dr RY Tsien 47 , respectively. All constructs were confirmed with fluorescence-based automated sequencing. cDNAs were subcloned into the high-expression pGEMHE vector 48 for expression in Xenopus oocytes. Vectors were linearized and complementary RNA (cRNA) was transcribed using the mMessage mMachine kit (Ambion). For co-injection experiments, relative amounts of RNA were quantified on agarose gels with a densitometric method. CNGA1 and CNGB1 were injected at an RNA ratio of 1:4 that has previously been The CnGA1 subunits (red) initially assemble into trimers through homotypic association of their carboxy-terminal CLZ domains into a parallel three helix coiled coil. subsequently, a CnGB1 subunit (green), if present, is preferentially incorporated into the remaining slot of the channel tetramer due to a high affinity association with CnGA1 subunits. (b) Deletion of the CLZ domain in CnGA1 allows CnGA1 subunits to assemble with CnGB1 subunits. This produces channels with multiple CnGB1 subunits and a large dominant negative effect of CnGB1. or the sum of two Langmuir isotherms: -diltiazem cis where K 1/2 is the apparent affinity and X is the fraction of the high affinity component (K a 1/2 ) relative to the low affinity component (K b 1/2 ). For FRET experiments, fluorescence images were collected using a X20 objective from the animal pole of oocytes with a confocal microscope (Leica SP1) using laser excitation at 458 nm and 488 nm. Regions of interest were drawn by hand in MetaMorph (Molecular Devices), and background fluorescence was quantified from the blank area inside the cell and subtracted ( Supplementary Fig. S6 ). The pinhole was adjusted so that the brightest cells were within the linear range of the detector with a maximum pinhole setting of 1 Airy unit. Data from cells with very dim fluorescent signals were discarded. Emission spectra were constructed over a range of wavelengths with an emission window of 2 nm. Spectra closely resembled the published spectra 46, 47 , suggesting that the fluorescent properties of these fluorescent protein variants are retained despite their incorporation into fusion proteins. FRET calculations were made using the spectral FRET method, as previously described 7, 9, 23 . Protein biochemistry. Four different CLZ containing constructs of varying lengths were generated for both CNGA1 and CNGA3 ( Supplementary Fig. 3) . Each of these eight constructs was subcloned into two bacterial expression vectors, pNGFP_BC and pCGFP_BC (kindly provided by Dr. E. Gouaux), creating aminoterminal and carboxy-terminal GFP fusion proteins respectively. Each of these 16 constructs was screened by FSEC as previously described 28 . Briefly, 5 ml bacterial cultures were induced, spun down, and resuspended in 800 µl of 150 mM KCl, 30 mM HEPES, pH 7.5 containing 2.5 µg ml − 1 DNAse and cOmplete protease inhibitor tablets (Roche). The cells were lysed with a probe sonicator; the lysate was cleared by centrifugation (1,700 × g for 30 min); and 100 µl of the supernatant was loaded on a Superdex 200 10/300 GL column (GE Healthcare) mounted on an FPLC system with a fluorescence detector set for detection of GFP fluorescence. Four candidates from the FSEC screen (CNGA1#621-690, CNGA1#621-667, CNGA3#626-694, and CNGA3#626-672) were subjected to further analysis. For large scale protein preparations, the CNGA1 and CNGA3 channel fragments were subcloned into the pETGQ and pETM11 vectors, respectively, creating amino-terminal poly-His fusion proteins. For the CNGA3#626-672 construct, the NcoI site at the amino terminus was deleted using PCR-based methods. Two litre bacterial cultures were grown to mid-log phase and induced with 1 mM IPTG overnight at 18 °C. The cultures were spun down and resuspended in 150 mM KCl, 30 mM HEPES, pH 7.5 containing 1 mM AEBSF, 2.5 µg ml − 1 DNAse and cOmplete protease inhibitor tablets (Roche). Cells were lysed with an EmulsiFlex C-5 homogenizer (Avestin), and the lysate was cleared by centrifugation at 131,000 × g for 45 min at 4 °C. The proteins were then purified on a Ni 2 + affinity resin column (HisTrap HP, GE Healthcare). The octahistidine tag was removed by thrombin (1) (1) (2) (2) or TEV cleavage, and the proteins were further purified on an anion or cation exchange column (HiTrap Q HP or HiTrap SP FF, GE Healthcare) and concentrated. Two of the samples mentioned above (CNGA1#621-690, CNGA3#626-694) were submitted to the Biophysics Resource of the W.M. Keck Biotechnology Facility at Yale University (New Haven, CT) for LS-SEC analysis. The samples were run on a Superdex 200 10/300 GL column (GE Healthcare) in 150 mM NaCl, 20 µM EDTA, 50 mM Tris-HCL, pH 8.0 (CNGA1#621-690) or 150 mM KCl, 30 mM HEPES, pH 7.5 (CNGA3#626-694). The molecular weight was determined by solving the equation that relates the excess scattered light, measured at several angles, to the concentration of solute and the weight-average molar mass using the ASTRA program 30 . X-ray crystallography. Crystals were grown by the sitting drop, vapour diffusion method at 20°C. Crystals of CNGA1#621-690 grew using a 1:1 mixture of protein and reservoir solution containing 18.2% (w/v) PEG 3350, 45 mM Zn 2 + acetate, 9 mM CoCl 2 . These conditions produced crystals within 8 days with space group P2 1 2 1 2 1 that diffracted to a resolution of 2.14 Å. Crystals of CNGA3#626-672 grew using a 1:1 mixture of protein and reservoir solution containing 20% (w/v) PEG 3350, 200 mM K + acetate. These conditions produced crystals within 1 day with space group P12 1 1 that diffracted to a resolution of 1.9 Å. For diffraction data collection, crystals were immersed in liquid nitrogen after cryoprotection in 20% glycerol. Data were collected at 110 °K on beamline 8.2.1 at the Advanced Light Source (Lawrence Berkeley National Laboratory, Berkeley). Integration, scaling and merging of the diffraction data were done with the Mosflm program 52 . The structures were solved by molecular replacement using the programs Phaser 53 and Phenix 54 . For CNGA1, the structure of HIV gp41 N-trimer pocket region 33 (3L36) was used as a search probe. For CNGA3, the structure of SARS virus S2 protein 34 (1ZVB) was used as a search probe. Molecular models were rebuilt using ARP/wARP (ref. 55) (CNGA1) or the Phenix software suite 54 (CNGA3). Structure refinement and validation were performed using the Phenix software suite 54 and the program Coot 56 using Fo -Fc ′omit′ maps. Based on the Ramachandran plot, the percent of amino acids in favoured/allowed/ disallowed conformations were 99.37/0.63/0.0 for CNGA1 and 100.00/0.0/0.0 for CNGA3. The coiled-coil parameters were analysed by the program TWISTER 57 . Statistical analysis. Data parameters were plotted as mean ± s.e.m. Student's test was used to determine significance at P < 0.05.

The 2011 Retrovirology Prize winner Masao Matsuoka: forward looking and antisense

Masao Matsuoka wins the 2011 Retrovirology Prize.

LTR DNA sequence of the HTLV-1 provirus is preferentially methylated while the 3' LTR of the provirus is not. His work in this area is independent of and contemporaneous with similar observations made in human immunodeficiency virus type 1 (HIV-1) research and has added to our insights on how retroviral latency is achieved and how it might be thwarted by employing targeted therapeutic agents. Because Matsuoka found that the 3' LTR of the HTLV-1 provirus is surprisingly hypomethylated, this led him to consider whether such a state would favor an antisense transcript originating from this end of the proviral genome. HIV-1 researchers have long ago discounted the importance of a 3' LTR driven antisense transcript. Yet, in the last decade, independent results from Mesnard (France) and Matsuoka (Japan) have strongly established the existence and biological importance of an HTLV-1 antisense transcript called HBZ. The HBZ RNA encoded by the minus strand of the provirus is capable of encoding a bZIP protein, and Matsuoka and his colleagues have shown that this RNA and its protein are expressed in all Adult T-cell leukemias (ATLs). Importantly, Matsuoka is the first to demonstrate that the expression of HBZ promotes the proliferation of T cells, and he has reported that HBZ expressing transgenic mice develop T cell lymphomas. Professor Matsuoka has been a founding member of Retrovirology's editorial board. To understand better Masao's views, I asked for his answers to several questions. KTJ: Tell me what you find the most rewarding about being a retrovirologist? MM: Human retroviruses are big threats to humans. They evolve to replicate elaborately in human cells using their small genomes. Whenever I find clever mechanisms of retroviruses, they never failed to surprise Correspondence: kjeang@niaid.nih.gov Laboratory of Molecular Microbiology, National Institute of Allergy and Infectious Diseases, the National Institutes of Health, 9000 Rockville Pike, Bethesda 20892, MD, USA me. At the same time, I feel the challenge to eliminate them from humans and to prevent retroviral diseases. As a retrovirologist, I am fascinated to study these small, but immensely important genomes with their ingenious mechanisms. KTJ: Did you ever have second thoughts about being a scientist? What would you have done if you did not become a scientist? MM: When I was a boy, I was fascinated with living fish and insects. I was very much interested in living animals. It is a reason that I love fishing. If I did not become a scientist, I would have liked to be a fisherman. However, for Japanese fishermen, it is very competitive to catch good fish for "Sushi". I am skeptical that I can succeed as a fisherman. I guess that being a scientist is better for me. KTJ: The 21 st century is said by many to be the "Pacific century"how do you see science being different in the 21 st century compared to the 20 th century? MM: I am confident in the 21 st century that people in the Pacific countries will be influential in science. A striking feature of this region is diversity in many aspects, including race, culture, politics, and so on. I hope that this huge diversity generates new science, and contributes to human society to solve many challenging puzzles like climate changes, and ever changing diseases (diabetes mellitus, cancer, infectious diseases, etc.), and the exhaustion of natural resources. Of course, none of these issues are easy to be solved. I really hope that diverse people in Pacific countries contribute to these tough problems. KTJ: Young scientists face many obstacles today in developing their careerswhat do you see as some obstacles facing young Japanese scientists and how would you advise them? MM: In most Japanese universities, associate and assistant professors work under the supervision of full professors. Full professors get grant money from the government or funding agencies, and they usually determine the research theme for young scientists. It is not an actual independent position. I would like to recommend young scientists to be independent as possible as they can or to choose a good boss who permits them to do independent research. KTJ: Someone told me once that every 10 years he reinvents himself. What do you see yourself doing 10 years from now that might be different from what you are doing today? MM: I agree that reinventing oneself is a nice and cool idea. However, I think that people (except rare persons) tend to behave as they always are. Before I moved to this Institute, I was a clinician, and took care of many patients in the hospital. I spent small efforts for my research at that time (half clinician, and half scientist). However, now I spend much more time for my research in Institute for Virus Research, Kyoto University. My past background is good experience for me since I can appreciate different points of view for retrovirology. Actually, I think understanding how the virus induces various clinical conditions (inflammation, hypercalcemia, immunodeficiency, etc.) from virological and clinical points of view provide informed and synergistic perspectives. Sometimes, these different views lead to new ideas. KTJ: You have published 6 papers [4] [5] [6] [7] [8] [9] in Retrovirologytell me what do you think about Open Accessshould this principle be important to Japanese scientists, why or why not? MM: I strongly agree with Open Access. Access to knowledge should be open to anyone who wants it. Recently, Japanese government and funding agencies do not request open access to scientists. This policy should be changed. Kyoto University has its own repository (Kyoto University Research Information Repository), Kurenai (that means "truly red" in Japanese), to access publications from Kyoto University. The world is sometimes unfair. But, science should be open and fair for anyone who wants to know new findings and knowledge. It should be the concrete basis for science. KTJ: When the time comes and you retire from doing science, how and for what would you like to be remembered? MM: When I retire from science, I would like to be remembered as a researcher who connected clinical science and retrovirology. Of course, I also study drug development for HIV-1 and resistance mechanisms. I am confident that there are many other important contributors to this field of HIV. KTJ: Finally, if you have an opportunity to address a gathering of all the world leaders at the United Nation, in 250 words or lesswhat would you say to them? MM: Now, the world has become closely connected more than at any other time of human history in many aspects. Economies are tightly linked, and an economical crisis like the Greek crisis in one country quickly affects many other world economies. Climate changes also influence the world. In the latter, a limitless desire of human exploitation of our environment has exacerbated these situations. A new viral infection quickly spreads within weeks like SARS. The HIV pandemic is a big tragedy to human, and avian influenza will be a new threat. To respond to these infections, we need a new platform of collaboration and friendship. Thus, we must appreciate that people and countries are indivisibly united. However, the sad thing is that people of the world are stuck in old thinking. We need to understand that in today's world it is not possible for one country to pursue mainly its own benefit. The "chyu-you (derived from the famous Chinese book)" means the balanced state or person with harmonized character, which is appreciated by people for a long time in Japan. I hope that people and countries should keep this word in mind, and exercise control over themselve. The host and the pathogen co-evolve to adapt to each other. It is time for humans to harmonize the world and society. Science is an important arena which gives us many opportunities to achieve this.

Effect modification of environmental factors on influenza-associated mortality: a time-series study in two Chinese cities

BACKGROUND: Environmental factors have been associated with transmission and survival of influenza viruses but no studies have ever explored the role of environmental factors on severity of influenza infection. METHODS: We applied a Poisson regression model to the mortality data of two Chinese metropolitan cities located within the subtropical zone, to calculate the influenza associated excess mortality risks during the periods with different levels of temperature and humidity. RESULTS: The results showed that high absolute humidity (measured by vapor pressure) was significantly (p < 0.05) associated with increased risks of all-cause and cardiorespiratory deaths, but not with increased risks of pneumonia and influenza deaths. The association between absolute humidity and mortality risks was found consistent among the two cities. An increasing pattern of influenza associated mortality risks was also found across the strata of low to high relative humidity, but the results were less consistent for temperature. CONCLUSIONS: These findings highlight the need for people with chronic cardiovascular and respiratory diseases to take extra caution against influenza during hot and humid days in the subtropics and tropics.

Influenza used to be considered as a "cold" disease as it usually returns every cold winter in temperate countries. However, recent studies have shown that influenza can be active throughout the year in the warm tropics and subtropics and the disease burden of influenza there can be as heavy as that in temperate climates [1] [2] [3] [4] . It has been proposed that influenza seasonality is driven by complicated interactions between antigenic drifts of virus strains, environmental factors, host susceptibility and behavior changes [5] [6] [7] . Among these potential factors, environmental factors including temperature and relative humidity have been most thoroughly explored by laboratory and observational studies [8, 9] . Recent studies raised a hypothesis that absolute humidity is one of drivers for influenza seasonality in temperate regions [10, 11] , but the mechanism behind various seasonal patterns of influenza outbreaks under different climates remains unclear. Given the association of environmental factors with both influenza virus activity and mortality, these factors have been adjusted for as confounders in the statistical models for influenza associated mortality burden [2, 12] . However, few of previous laboratory or epidemiological studies tackled the potential modifying role of environmental factors on severity of infection, or mortality risks associated with influenza. Such effect modification is plausible because extreme weather conditions could have a synergistic effect with influenza on mortality burden. Both cold and hot temperatures have been associated with increased mortality risks of respiratory diseases in numerous studies [13] [14] [15] . Our previous study in Hong Kong also found a two-peak seasonal variation in mortality burden of influenza, which is similar to the pattern of influenza seasonality, suggesting that environmental factors might also affect the severity of seasonal influenza infection [16] . In this study we applied a Poisson regression model to the data of two subtropical Chinese cities: Guangzhou and Hong Kong, to examine the possible effect modification of environmental factors on the severity of influenza infection measured by the mortality attributable to influenza. These two cities are geographically close, with Guangzhou located at latitude 23°N and Hong Kong at 21°N ( Figure 1 ). Both cities have a typical subtropical climate, but on average Hong Kong has a higher temperature and humidity. Hong Kong also has a larger population than Guangzhou (6.8 million vs. 3.7 million).We examined three environmental factors which have been documented to regulate virus survival and transmission: temperature, relative humidity and absolute humidity. According to Hong Kong Law all deaths from natural causes are required to be registered, so the mortality data in Hong Kong was fairly complete. The registered death data collected by the Guangzhou Department of Health covered the residents from the eight urban districts, but did not include the immigrant population and residents living in the two suburban districts. Therefore, the population denominator of Guangzhou did not change much during the study period. Given the fact that most immigrant workers are young adults and few died in Guangzhou, we can assume that our mortality data was complete and representative of the general population of Guangzhou. We aggregated the weekly numbers of deaths with underlying cause of cardiorespiratory disease, pneumonia and influenza, and allcause mortality. The corresponding International Classification of Diseases, Tenth Revision (ICD10) codes adopted by Guangzhou and Hong Kong are I00-I99, J00-J99 for cardiorespiratory, J10-J18 for pneumonia and influenza, and A00-R99 for all-cause deaths. We used the accidental deaths (ICD10 codes S00-T989) in Hong Kong as the control disease. Weekly mean temperature and relative humidity were separately derived from the National Meteorological Information Centre of China and Hong Kong Observatory for Guangzhou and Hong Kong respectively. Weekly mean vapor pressure was calculated as a metric for absolute humidity following an equation provided by Basu et al. [18] , vaporpressure = 0.06112*relativehumidity% * 10 7.5*temperature( • C)/(237.7+temperature( • C)) (1) Vapor pressure will be used for absolute humidity hereafter in this paper. A generalized additive model (GAM) with a log link function and Poisson error was fitted to the weekly numbers of deaths. The Poisson model has been widely applied to the estimation of influenza associated disease burden and recently been validated by an empirical dataset of laboratory confirmed influenza cases [19] . First, long-term trends and seasonal patterns of causespecific mortality counts, as well as environmental factors including temperature and relative humidity, were adjusted for as confounders by building a core model: Here y t denotes the number of deaths at week t. ns(t), ns(temp t ), and ns(humd t ) denote the natural cubic spline smoothing functions of time, weekly average temperature and relative humidity. The natural spline smoothing function was used to remove the small variation while maintaining the major trend of each variable, i.e. to make them smoother. The aim of smoothing was to increase the efficiency in estimating the model coefficients [20] . Given the high correlation between temperature and vapor pressure, the smoothing function of vapor pressure was not added in order to avoid collinearity between the variables. The adequacy of this core model was evaluated by the absence of any obvious pattern in partial autocorrelation functions of its residuals (Additional file 1: Figure S1 ). Second, the weekly proportions of specimens positive for influenza A or B were then added into the core model as a variable for influenza virus activity to obtain a main effect model. The main effects of influenza have been presented elsewhere [21] . To explore the effect modification of environmental factors on influenza associated mortality, we added into the core model the product terms of influenza proportion variable and dummy variables for periods of normal and extreme (high or low) weather conditions as interaction terms between virus activity and environmental factors. For example, the interaction model for temperature and influenza mortality was: where Lowtemp i = 1 for the periods within low temperature ranges and 0 for otherwise, and Midtemp t and Hightemp t are similarly defined as the dummy variables for the middle and high temperature periods. The smoothing function of temperature ns(temp t ) was kept in the model in order to adjust for the association of temperature and mortality. It is reasonable to assume that the cutoff point of extreme weather may differ across cities as people may adapt to prevailing climates. We used the first (25%) and third quartiles (75%) of weekly average temperatures (or humidity) as the cutoff points to define the low, middle and high temperature (or humidity) periods [22] . The presence of effect modification by environmental factors was evaluated using likelihood ratio tests between the interaction and main effect models. To measure the effects of influenza on mortality, we computed the percentage change of mortality counts associated with 1% increases of influenza virus activity for the low, middle and high periods. The formula for the low temperature period is Here β 1 was obtained from the above interaction model. To test whether our results were robust to various definitions of periods, we chose two sets of extra cutoff points: 20th and 80th, 30th and 70th percentiles of weekly average data in each city. Although we removed the autocorrelation and seasonal trends within mortality data, there are still concerns that such adjustment was inadequate and the remaining uncontrolled seasonal factors may cause interaction terms of environmental factors and virus activity to appear significant in our models. To rule out this possibility, we used accidental deaths that were expected to be unrelated to influenza infection as a control mortality group to show that our findings are unlikely the spurious results of under-adjustment of seasonal confounding factors in modeling. Some previous studies used the anomalies to assess the effects of meteorology factors on influenza [11] .We therefore conducted a sensitivity analysis by defining the strata by anomalies, instead of absolute values of meteorology factors. The anomalies were defined as the deviations of observed metrological data from a seasonal curve with a constant and a sinusoidal pair fitted to these observed data. All the analyses were performed using the mgcv package of R software (version 2.5.1.) [23] . Hong Kong has a larger population than Guangzhou (6.8 million versus 3.7 million) during our study period. On average, Hong Kong has a slightly higher temperature, relative humidity and vapor pressure, and smaller annual variations than Guangzhou ( Table 1 ). The mean of the weekly proportions of specimens positive for influenza A or B was higher in Hong Kong than in Guangzhou. In 2004 and 2006, there were two peaks of virus activity in Hong Kong (one in February/March and another in June/July), but only one broad peak in Guangzhou ( Figure 2 ). The influenza seasonality was similar between these two cities in 2005. More deaths with underlying causes of cardiorespiratory, pneumonia and influenza or all-cause were recorded in the low temperature (or low vapor pressure) periods for both Guangzhou and Hong Kong ( Table 2) . For both cities, the mortality counts did not show any obvious difference across levels of relative humidity. High influenza virus activity coincided with high levels of temperature, relative humidity or vapor pressure, with the only exception of Guangzhou which had higher mean proportions when temperature is within the middle range. Significant interaction between temperature and influenza on the mortality risks was only found in cardiorespiratory mortality in Hong Kong (p < 0.05), and the interaction between relative humidity and influenza was found significant for all-cause mortality in both Guangzhou and Hong Kong, and for cardiorespiratory mortality only in Guangzhou ( Table 3 ). The patterns of influenza impact across the low to high temperature periods were not consistent among the two cities. In Guangzhou, the highest risk of mortality tended to be observed in the middle-temperature period, whereas for Hong Kong the largest changes in risk were found in the periods with high temperature for the three mortality categories ( Table 3 ). An increasing pattern of influenza associated mortality risks could be observed along low-, mid-and high-relative humidity periods in Guangzhou and Hong Kong, but most of estimates for low-and high-periods were not statistically significant ( Table 3) . The interaction between vapor pressure levels and virus activity were found to be significant (p < 0.05) for allcause and cardiorespiratory mortality, but not for pneumonia and influenza mortality (Table 3) . Consistently higher mortality risks were found at the high levels of vapor pressure, with only exception of P&I mortality in Hong Kong. The influenza effects on mortality of cardiorespiratory, pneumonia and influenza, and all causes were found significant during the middle-and highvapor pressure periods (with the only exception of allcause mortality in Hong Kong), but not significant when vapor pressures remained at the relatively low levels. For the high vapor pressure periods, all-cause excess mortality counts attributable to influenza would increase by 0.35% and 0.26% for per 1% increase of virus activity, and the corresponding increases for cardiorespiratory mortality were 0.54% and 0.49% for Guangzhou and Hong Kong, respectively (Table 3) . We also assessed the effect modification of environmental factors on influenza effects (i.e. interaction of each factor and influenza) for the age group younger than 65 years (< 65) and the elderly aged 65 years or older (≥65). The results of the ≥65 age group were consistent with those for the all-ages group, with an increasing trend over vapor pressure levels observed for all-cause and cardiorespiratory mortality in the two cities (Figure 3 ). For the < 65 age group, this trend could also be observed in most city-specific disease categories, with the only exception of all-cause mortality in Hong Kong. But the interaction terms were statistically significant (p < 0.05) only in the ≥65 group. The modifying effects of temperature and relative humidity on influenza effects were found quite similar between the all-ages and ≥65 age groups, but not between the allages and < 65 age groups (data not shown). The models with indicators for different cutoff points returned similar estimates for temperature, relative humidity and vapor pressure (Additional file 1: Figure S2 ). We did not observe any significant interaction between influenza and environmental factors in terms of their effects on the control mortality category of accidental mortality in Hong Kong. Influenza associated accidental mortality risks were also not statistically significant. To ensure the standardized comparison between Guangzhou and Hong Kong, we decided to apply the same modeling approach to the data of same study period, because the core model would have been slightly different if we used the longer time series of Hong Kong. To check the robustness of our conclusions, we repeated the above analysis using a longer time series data of Hong Kong during 1998-2006 and the estimates were shown in Additional file 1: Table S1 . The statistical significance of interaction terms was close to those from the study period of 2004-2006. The estimates for the low temperature periods became larger and statistically significant; those for the high relative humidity periods were smaller and comparable to the middle relative humidity; and for the low vapor pressure periods, the estimates were similar but became significant. The increasing trend across the low, middle and high levels of temperature and relative humidity was less evident. The results of stratification analysis by anomalies are shown in Additional file 1: Table S2 . The estimates were similar to those for the periods defined by absolute values of meteorological factors, in terms of magnitude and changing patterns. But the likelihood ratio tests showed more significant interaction for temperature or vapor pressure, and less for relative humidity. In this study we quantified influenza associated mortality risks at the various ranges of temperature and humidity, and compared the results between two large subtropical cities. An increasing pattern of influenza-associated mortality risks on all-cause and cardiorespiratory along the low to high periods was found for temperature, relative and absolute humidity in both cities during the period of 2004-2006. The interaction between vapor pressure indicators and virus activity were also consistently significant for relative and absolute humidity. Although lower vapor pressure (or temperature) has been found to facilitate virus transmission and survival in the guinea pig model [10] , our results suggested that higher vapor pressure (or temperature) was associated with a higher mortality risk attributable to influenza. Severity of seasonal influenza epidemics were not only determined by virus transmission efficiency and outdoor weather conditions, but also largely affected by host resistance, indoor living environment and social behavior [5] . The high vapor pressure periods coincided with the summer peaks of influenza in both Guangzhou and Hong Kong, indicating that influenza could pose a higher risk when reaching its peak in seasons with high vapor pressure in subtropical cities. Our results may help to interpret our previous findings that the effects of influenza were significantly higher in the humid and warm spring/summer period than in the dry and cold winter period in Hong Kong [16] . The extreme low vapor pressure was usually recorded during December-January and the highest appeared during June-July, which coincided with the trough and peak periods of excess risks associated with influenza viruses. Effect modification of environmental factors was only detected in all-cause and cardiorespiratory mortality, but not in pneumonia and influenza, suggesting that the synergistic interaction between high humidity (or temperature) and virus activity may mainly lie in their similar regulation pathways in cardiovascular systems. These results are also in agreement with our previous findings that pneumonia and influenza mortality risks attributable to influenza did not exhibit a seasonal variation [16] . Extreme heat has been documented to increase blood viscosity through evaporation of body fluid and trigger intravascular coagulation through damaging endothelial cells [24] . Influenza infection has a similar pro-thrombotic effect by inducing inflammation around blood vessels and rupturing atherosclerotic plaques [25] . As a result, mortality risks would be dramatically raised by the stress of both extreme weather and influenza infections. The results suggested that we need extra precautionary measures to reduce influenza infections in people with cardiovascular diseases, especially under the frequent hot and humid weather conditions experienced in the tropical and subtropical areas. Although the experiments of influenza virus transmission between guinea pig hosts found the higher transmission rates occurred under dry air (vapor pressure below 10hPA) [10] , our results indicated that the mortality risks associated with influenza under the low vapor pressure environment were lower than the rest of study period. The reason could be the short time of exposure to the very low level of vapor pressure. During our study period there were only 2 and 13 weeks with an average vapor pressure below 10hPA in Hong Kong and Guangzhou, respectively. Most weekly average vapor Percentage change (%) of mortality counts for all-cause and cardiorespiratory (CRD) mortality. The risks associated with 1% increase in influenza virus activity during the low-, middle-and high-vapor pressure periods were plotted for the age groups younger than 65 years (< 65) and equal or over 65 years (≥65). The 95% confidence intervals were shown in vertical bars. An asterisk is added above the bars if the interaction between vapor pressure and influenza is shown statistically significant by the likelihood ratio test between main effects and interaction models. pressures during the low vapor pressure period were within the range of 10-20hPA, in which the guinea pig experiments showed dramatically reduced transmission rates [10] . In future, we may examine the seasonal variation in influenza effects in other cities to assess whether such a seasonal variation, if common in subtropical and tropical cities, is consistently determined by environmental factors, or by other factors such as host immunity and virus virulence. The results for the different age groups suggested that the modification effects of environmental factors may mainly lie in the elderly aged over 65 years, as the consistent increasing trend over the low to high vapor pressure periods was only observed in this age group. However, since over 65% of deaths occurred in this age group for both cities (Table 1 ), the small numbers of weekly death counts in the younger age group (< 65 years) may not have had enough power to allow assessment of effect modification based on the data over the 3 years. A future study with a long study period or a large population may help answer whether young people maybe also expose to higher mortality risks during the hot and humid days. In this study, we used the quartiles of weekly data in each city, to separately define the periods with normal (middle) and extreme (low and high) weather. Given the difference in weather conditions between these cities, we think that it is not appropriate to use the same cutoff points for temperature or humidity to compare their modification effects on influenza associated mortality, as people living in hot subtropical and tropical regions may adapt well to the year-round hot and humid climate and have a higher threshold for adverse effects of weather. For example, although it is widely accepted that the temperature effects on mortality exhibited a U-or V-shape curve in both temperate and tropical/subtropical areas, the turning point of this curve varied across different cities. A study conducted in 11 cities of the US found that the turning point of temperature for its effects on mortality could range from 18.4°C to 32.4°C [26] . To our best knowledge, so far there are no studies that have ever assessed the effect modification of temperature on influenza effects. Therefore, the commonly adopted cutoff points of city-specific quartiles seem appropriate at this stage [21, 27] . There are several limitations in our study. Firstly, our study is based on 3 years of surveillance data which may not have enough power to allow assessment of exposureresponse curves for the effects of environmental factors. Nevertheless, our findings did suggest an increasing trend of influenza associated mortality risks across the periods of low, middle and high vapor pressure, although such findings may be applicable only to the warm climates. Secondly, we only investigated the effect modification of environmental factors through a simple interaction model, but there were other unadjusted factors, including host susceptibility and virulence of influenza strains. These factors are unlikely to work independently with environmental factors. Other environmental factors such as ultraviolet radiation [28] , rainfall [29] have been proposed to play a role in the regulation of influenza seasonality, although evidence is rather limited compared with the three factors we chose to investigate [30] . Lastly, we did not adjust for the vaccination rate in our model. In 2003, vaccination rate was 191 doses per 1,000 total population in Hong Kong [31] , slightly higher than the rate of 129 doses/1,000 total population in Guangzhou [21] . However, it is not clear when people received vaccination; therefore we were unable to assess the role of vaccination in our study. This study provides a piece of key evidence to the effect of environmental factors on severity of seasonal influenza under warm climates and helps reveal the mechanism behind global influenza seasonality. It also highlights the need for people with chronic cardiovascular and respiratory conditions to take extra caution against influenza during the hot and humid days in the subtropics. Additional file 1: Tables S1 and S2; Figures S1 and S2.

Influenza A/H1N1 septic shock in a patient with systemic lupus erythematosus. A case report

BACKGROUND: Immunocompromised patients, such as systemic lupus erythematosus (SLE) sufferers have an increased risk of mortality, following influenza infection. In the recent pandemic, influenza A H1NI virus caused 18449 deaths, mainly because of adult respiratory distress syndrome or bacterial co-infections. CASE PRESENTATION: In this case report, an SLE patient with viral-induced septic shock, without overt pulmonary involvement, is discussed. The patient was administered oseltamivir and supportive treatment, including wide-spectrum antibiotics, vasopressors and steroids, according to the guidelines proposed for bacterial sepsis and septic shock. She finally survived and experienced a lupus flare soon after intensive care unit (ICU) discharge. CONCLUSIONS: To our knowledge, this is the first case to report severe septic shock from influenza A/H1N1 virus, without overt pulmonary involvement.

Infections are among the most important causes of morbidity and mortality in systemic lupus erythematosus (SLE). However, viruses are not considered to cause serious infections in these patients; they, usually, represent reactivation of herpes viruses, such as herpes simplex virus and varicella-zoster virus [1] . Nevertheless, it is reported that immunocompromised patients have an increased risk of mortality, following influenza infection [2] . In the recent pandemic, influenza A H1N1 virus has been estimated to cause approximately 18.449 deaths in 214 different countries until August 1 st 2010 [3] . Adult respiratory distress syndrome (ARDS), along with bacterial co-infections were the direct causes of death in most cases [4] . However, no cases of viral-induced septic shock without severe pulmonary involvement have been reported. Nevertheless, little is known about this infection in SLE patients [5] . Herein, we report a case of an SLE patient, who developed septic shock due to influenza A H1N1 infection, without acute lung injury. A 46-year old female was admitted to the hospital because of low-grade fever, sore throat and fatigue for four days; she was on clarithromycin 500 mg twice daily, as she was considered to suffer from upper respiratory tract infection by her general practitioner. The patient had a history of SLE for 24 years, antiphospholipid syndrome and autoimmune hypothyroidism. SLE was diagnosed in the background of immune thrombocytopenic purpura (ITP), starting at the age of 8. At the age of 12, splenectomy was performed to control refractory thrombocytopenia. Currently, SLE was adequately controlled (Systemic Lupus Erythematosus Disease Activity Index, SLEDAI = 0, anti-dsDNA antibodies negative, C3 and C4 levels normal); medication included methylprednisolone 8 mg/day, azathioprine 50 mg/day, aspirin 50 mg/day and levothyroxine 150 μg/day. The patient had been vaccinated against seasonal influenza and Streptococcus pneumoniae a month before, but not against A/H1N1 virus. On admission, she was in severe cardiovascular instability with hypotension (BP = 60/40 mmHg), tachycardia (HR = 130/min), tachypnea (RR = 30/min), hypothermia (< 35.5°C), along with oliguria and altered mental status. Oxygenation fraction PaO 2 /FiO 2 was over 350. No obvious site of infection could be identified. The initial chest X-ray was normal, as well as computed tomography of the thorax ( Figure 1 ). Heart ultrasound revealed mild diastolic dysfunction with preserved ejection fraction and no valvular disease. Concerning other organ involvement, there was mild prerenal azotemia (urea 94 mg/dl, creatinine 1.5 mg/dl) and severe liver impairment (alanine aminotransferase 2837 U/L, aspartate aminotransferase 3165 U/L). The patient was considered to suffer from systemic inflammatory response syndrome (SIRS) and was treated with vigorous fluid resuscitation (crystalloids and colloids), oseltamivir 150 mg/day and moxifloxacin (400 mg/day, after appropriate cultures were obtained). A few hours later, she was intubated and carried to the ICU, because of refractory shock, where vasopressor therapy (noradrenaline up to 2 μg/kg/min) was administered to maintain mean arterial pressure ≥65 mmHg. Arterial pressure wave form analysis (FloTrac/Vigileo System), continuous ScVO 2 and CVP monitoring were used to estimate patient's hemodynamics. After initial fluid resuscitation, cardiac index, ScVO 2 and CVP measurements were 2.7 L/min/m 2 , 75% and 13 mmHg, respectively (mean values). Blood lactate levels were 4.8 mmol/L. The ventilatory support consisted of controlled MV with tidal volume 7 ml/kg and respiratory rate 15/min. Initial plateau pressure was 16 cmH 2 O with respiratory system compliance 58 ml/cmH 2 O. The patient responded after 24 hours; noradrenaline was tapered to 0.25 μg/kg/min and hydrocortisone (300 mg/ day) was added as adjunctive therapy. Additional antibiotics included piperacillin/tazobactam (18 g/day) and linezolide (1.2 g/day). Blood, urine and bronco-alveolar lavage (BAL) cultures were negative. Real-time PCR for influenza A/H1N1 virus (BAL specimen) was positive and oseltamivir was administered at 300 mg/day. In ICU, the clinical course was complicated by bacterial co-infections 10 days after admission; septicaemia due to carbapenem-resistant Klebsiella pneumoniae and ventilator-associated pneumonia due to multi-drug resistant Acinetobacter baumanii. Gentamicin and colistin were administered, according to strain sensitivity, leading to complete resolution of the lesions. She was extubated after 19 days and transferred to the ward. Concerning H1N1 virus, BAL rt-PCR was persistently positive for 21 days. Oseltamivir was administered in high doses (300 mg/day) for 28 days and was discontinued after negative PCR. The patient experienced a lupus flare (SLEDAI = 4, C3 = 65 mg/dl, C4 = 7.4 mg/dl), soon after extubation with thrombocytopenia and severe haemolytic anemia. The flare was successfully managed with steroids and intravenous immunoglobulins (IVIGs). Her clinical condition was complicated by severe critical illness polyneuropathy/myopathy (CIP/CIM), managed with long term physiotherapy. In four months, previous therapeutic regimen was re-established and satisfactory performance status was maintained. Influenza A/H1N1 pandemic represents a global health issue, as it was estimated to be the direct cause of over 600 million respiratory infections and over 50 million hospitalizations, until August 10 th , when WHO announced that H1N1 was in post-pandemic period [6]. Immunosupression is a well-defined risk factor for worse outcome in H1N1 disease. ARDS and secondary bacterial super-infections, leading to septic shock and multiple organ failure are considered to be the primary causes of death [4, 7] . However, primary viral septic shock, especially by influenza, is rarely reported in the literature [8] [9] [10] [11] . The mortality pattern in SLE is reported to be biphasic; major infections play an important role during the first years of disease, while cardiovascular and other disease complications account for most deaths in long-lasting disease [12] . Viral infections, however, are not reported to affect mortality in SLE [1, 13] . Severe cardiovascular instability, as indicated by the hemodynamic monitoring of the patient, with refractory shock in H1N1 infection, has not been reported so far in the literature. Acute lung injury and/or ARDS due to H1N1 virus could not be identified, as initial chest Xray and thorax CT were normal and oxygenation was not impaired (PaO 2 /FiO 2 ≥350) throughout the disease course. Bacterial infections, as potential causes of septic shock, were not diagnosed; serial blood, urine and BAL cultures were negative. The therapeutic approach was that of conventional septic shock, according to 2008 guidelines and resulted in patient recovery [14] . Long lasting viral persistence, despite recommended oseltamivir therapy, probably reflects a poor immune status and defective natural and acquired antibacterial immunity, due to several reasons. Patients with SLE are considered to be immunocompromised either because of the disease itself or due to the immunomodulating agents used for disease management [15] . On the other hand, splenectomy is not considered to represent a major risk factor for viral infections. Vaccination is suggested to be less effective in SLE, especially in the background of azathioprine treatment, although measurement of antibody titers can only indirectly assess the efficacy of vaccination [2] . Latest studies supported that seroprotection after a single vaccination in SLE patients was significantly reduced compared to healthy controls [16] . The presented patient was vaccinated against seasonal influenza but not against H1N1 virus. Although recent reports suggest that CD8+ T cells are able to cross react with H1N1 virus, they are functionally impaired [17] . The major complication in this patient, after ICU discharge, was a disease flare (thrombocytopenia and hemolytic anemia) along with CIP/CIM. The latter complicates approximately 30-50% of ICU patients, particularly in cases of multiple organ failure and septic shock [18] . The pathophysiologic process behind CIP/CIM is not fully elucidated; reactive oxygen intermediates, drug toxicity, steroid therapy and poorly controlled hyperglycemia are considered to be critical predisposing factors [18] . Treatment options are not well validated, although intravenous immunoglobulins seem to have a beneficial effect [19] . IVIGs were administered to this patient for managing severe autoimmune haemolytic anemia and thrombocytopenia; their benefit in recovery from CIP/CIM can not be assessed directly. However, IVIGs allowed quick steroid tapering and, possibly, prevention of further nosocomial infections. To our knowledge, severe septic shock from influenza A/H1N1 virus, without overt pulmonary involvement, has not been reported in the literature. The authors of this case report of special interest from many points of view, feel it would be a positive step to share their experience with experts in the field. Physicians' awareness and prompt and aggressive supportive treatment are expected to optimize patient outcomes. Written informed consent was obtained from the patient for publication of this case report and any accompanying images. A copy of the written consent (in Greek) is available for review by the Editor-in-Chief of this Journal.

Timeliness of contact tracing among flight passengers for influenza A/H1N1 2009

BACKGROUND: During the initial containment phase of influenza A/H1N1 2009, close contacts of cases were traced to provide antiviral prophylaxis within 48 h after exposure and to alert them on signs of disease for early diagnosis and treatment. Passengers seated on the same row, two rows in front or behind a patient infectious for influenza, during a flight of ≥ 4 h were considered close contacts. This study evaluates the timeliness of flight-contact tracing (CT) as performed following national and international CT requests addressed to the Center of Infectious Disease Control (CIb/RIVM), and implemented by the Municipal Health Services of Schiphol Airport. METHODS: Elapsed days between date of flight arrival and the date passenger lists became available (contact details identified - CI) was used as proxy for timeliness of CT. In a retrospective study, dates of flight arrival, onset of illness, laboratory diagnosis, CT request and identification of contacts details through passenger lists, following CT requests to the RIVM for flights landed at Schiphol Airport were collected and analyzed. RESULTS: 24 requests for CT were identified. Three of these were declined as over 4 days had elapsed since flight arrival. In 17 out of 21 requests, contact details were obtained within 7 days after arrival (81%). The average delay between arrival and CI was 3,9 days (range 2-7), mainly caused by delay in diagnosis of the index patient after arrival (2,6 days). In four flights (19%), contacts were not identified or only after > 7 days. CI involving Dutch airlines was faster than non-Dutch airlines (P < 0,05). Passenger locator cards did not improve timeliness of CI. In only three flights contact details were identified within 2 days after arrival. CONCLUSION: CT for influenza A/H1N1 2009 among flight passengers was not successful for timely provision of prophylaxis. CT had little additional value for alerting passengers for disease symptoms, as this information already was provided during and after the flight. Public health authorities should take into account patient delays in seeking medical advise and laboratory confirmation in relation to maximum time to provide postexposure prophylaxis when deciding to install contact tracing measures. International standardization of CT guidelines is recommended.

Aircrafts can function as transport vehicle for patients infected with influenza, leading to introduction of a new virus strain to non-endemic areas [1, 2] . Although the risk is small, passengers might be infected by a contagious patient during the flight [3] [4] [5] [6] , as well as during public transport [7] . Transmission during the flight increases the possibility of further transmission in the area of destination. For these reasons, during the initial phase of the influenza A/H1N1 2009 pandemic, many countries initiated contact tracing among flight passengers of flights where contagious patients with laboratory confirmed influenza A/H1N1 2009 were notified. A risk assessment guideline for infectious diseases transmitted on aircrafts has been developed by the European Centre for Disease Prevention and Control (ECDC) [8] , which includes influenza. Literature study revealed on-board transmission in flights with a duration of less than 8 h. The majority of infected contacts during these flights were seated on the same row, or one or two rows in front of behind the index [9] [10] [11] [12] . Contacts up to 8 and 10 rows distance from the index were infected in one study [10] . As these contacts also had personal contact with the index during the flight, transmission across a distance of so many rows is not proven. The guideline concludes that it is difficult to design a single contact tracing algorithm for influenza. Due to the short incubation period of influenza, it is almost impossible to provide contacts with postexposure prophylaxis (PEP) within the time that it is most effective, which is 48 h after exposure [13] . Therefore, the main aim of contact tracing might be to interrupt the chain of transmission by alerting contacts for early diagnosis and treatment. Although the World Health Organization (WHO) developed technical advice for case management of influenza A/H1N1 2009 in air transport during the pandemic [14] , no international standardized protocol for contact tracing for this pathogen was available. In line with the ECDC guideline [8] and the Dutch guideline for 'Incidental introduction of a new influenza strain' [15] , in the Netherlands close contacts of a patient with laboratory confirmed pandemic influenza were identified. In case the index had been contagious during a flight with a duration of ≥ 4 h, passengers and cabin crew were to be informed on signs and symptoms of the disease and to seek medical care in case they would occur. In addition, close contacts, defined as passengers seated on the same row, two rows in front and two rows behind the index case, as well as the cabin crew working in this compartment, were traced by public health authorities to provide a 10 day prophylactic course of oseltamivir as soon as possible (preferably within 48 h after exposure). Schiphol Airport is the only airport in the Netherlands where trans-Atlantic flights arrive. Its Municipal Health Services (MHS, GGD Kennemerland) and the Center for Infectious disease Control (CIb-RIVM) frequently experienced that, despite all efforts, the time period elapsing from exposure to administration of the first oseltamivir dose exceeded the required 48 h. Acquiring contact details from airlines was time consuming, and contact details on passenger lists were often minimal, so that contacts were difficult to trace. In this study, we assess the time delay in contact tracing of flight passengers for influenza A/H1N1 2009 as performed in the Netherlands during the initial phase of the pandemic. Our data show that despite all efforts the effectiveness of this control measure in daily practice is minimal. From April 29th until June 22nd 2009, contact tracing among flight passengers in the Netherlands was indicated for laboratory confirmed influenza A/H1N1 2009 cases, who traveled on a flight for 4 h or longer while being contagious, defined as 1 day before, until 7 days after disease onset. These criteria were installed by the CIb, which also functions as National focal point (NFP). The procedure for contact tracing is complex, see Figure 1 . Requests for contact tracing to the CIb for Dutch index patients originate from any Dutch MHS which identifies a patient who traveled by plane while being contagious for an infectious disease which requires contact tracing. Other nation's health authorities will make a request to the CIb in case they diagnosed a patient which arrived at Schiphol airport for transit while being infectious. Requests for CT in the last group are submitted to the National Focal Point (NFP) or through the Early Warning and Response system of the EU (EWRS). The CIb verifies laboratory confirmation, and the indication for contact tracing regarding flight duration. The MHS of the airport where the specific flight landed coordinates contact tracing for flight passengers. In case of Schiphol, MHS Kennemerland approaches the involved airline company requesting the passenger list. The airline provides passenger lists with at least passenger names, seat numbers and booking or contact details. MHS Kennemerland then completes contact details through booking offices or using other search methods. Close contacts living in the Netherlands are traced by the respective Dutch MHS's. For tracing foreign contacts, the CIb sends a notification with contact details to the NFP of the country of final destination, or through the EWRS system for EU countries. During the pandemic, CT requests were turned down if more than 4 days had elapsed after flight arrival, as contact tracing was not considered to have additional value. During the study period, passenger locator cards (PLC) only were used on direct flights from Mexico during the initial phase of the pandemic. These flights were all run by Dutch airlines. For each contact investigation performed in the period April 29th until June 22nd 2009, the following data were collected: flight arrival date, first day of illness of index patient, date of laboratory diagnosis, date of contact tracing request and the date passenger lists were obtained and contact details were completed ('contacts details identified'). From these data, time intervals (in days) between flight arrival and date of diagnosis (interval I), between diagnosis and request dates (interval II) and between request and contact details identified dates (interval III) were calculated, see Figure 2 . Date of actual contact tracing and oseltamivir administration was not available in this study, but is inherently always hours if not days later. As the airline company traces contacts amongst crewmembers, these are not included in this study. Data were analyzed using SPSS software (version 18, USA). The influence of availability of PLC's on timeliness and the origin of the airline company (Dutch or non-Dutch) were statistically analyzed. In the period April 29th until June 22nd 2009, 24 indications for CT were identified. Three international requests concerning CT for influenza patients diagnosed outside the Netherlands were declined as already more than 4 days had elapsed since flight arrival. In 17 out of the 21 remaining contact investigations, passenger lists with contact details were obtained within 7 days after arrival (81%), see Table 1 . In total contact details of 451 close contacts were identified, of which 199 contacts lived in the Netherlands, and 252 contacts abroad. The average number of close contacts per flight was 27 (range: 8-44). In four contact investigations (19%), contact details were not obtained, or provided later than 7 days after flight arrival and CT was stopped. These CT were all related to non-Dutch airlines, and total delay *:. In the beginning of the pandemic one request for contact tracing was accepted after 7 days **: these late CT requests were accepted as the passenger lists of the concerned flights already were available from earlier contact investigations ***: date of diagnosis not known ª: Passenger Locator Card was stated 8 days for further data processing. Of the 21 requests, the total delay between request and contact detail identification was longer for non-Dutch airlines (mean 6,3 SD 2,7) compared with Dutch airlines (mean 4.1 days, SD 1,5)(1-sided Mann-Whitney test, p = 0,033). For the 17 completed contact investigations, interval I was the largest interval in the contact tracing procedure (mean 2,6 days, range 1-6, 95% CI 1,6-3,6, n = 13). The other intervals II and III were shorter, with a mean of 0,8 days and 0,6 days respectively, see Table 1 . Figure 3 shows the medians of the described intervals. Since 15/ 17 index cases were already ill before, or during the day of arrival of the flight, the delay in interval I is mainly caused by delay in seeking medical advice and diagnostic procedure itself. After acceptance of the request for CT by the CIb, GGD Kennemerland needed on average 0,6 days (range 0-2, 95% CI 0,3-0,9 days) to collect the passenger list from the airlines and complete contact details (interval III). The total delay between flight arrival and identification of contact details was on average 3,9 days (range 2-7 days, 95% confidence interval 3,2-4,7 days), see Table 1 . In only 3 out of 17 contact investigations (18%), contacts were identified within 2 days after arrival. In 2 out of these 3 contact investigations, PLC's were available. Interval III of the 5 CT with PLC's available was shorter (0,4 days, SD 0,5) than for 12 CT's without PLC (0,7 days, SD 0,7), this was not significant however (p: 0,25). Overall delay in CT with PLC's also was shorter (mean 3,6, SD 1,8), but not significant, when compared to CT without PLC's (mean 4,1, SD 1,4) (p:0,25). In this study we evaluated the timeliness of contact tracing (CT) of flight contacts in daily practice. We conclude that the prevailing policy to provide close contacts antiviral PEP during the early phase of the influenza pandemic is very difficult to implement effectively and therefore has little effect to control disease spread. Active case finding through contact tracing of exposed persons is an important procedure during the containment phase of an emerging communicable disease. However, our data show that, even in a small-industrialized country with modern communication tools, tracing of flight contacts exceeds the required maximum of 48 h after exposure. For influenza, close contacts of contagious index cases are entitled to receive antiviral PEP within 48 h after exposure to prevent them from becoming ill and further spreading of the disease. Starting oseltamivir within 48 h does not prevent disease but shortens the disease period, mitigates symptoms and might decrease further transmission. Awareness among contacts to seek medical evaluation when influenza-like (ILI) symptoms occur, for both proper antiviral treatment and (home-) isolation advice, reduces further spreading. As influenza has a relative short latent period, for influenza A(H1N1)/2009 varying between 0,7-3,1 days [16, 17] , contacts ideally should be informed within 1 day. Oseltamivir postexposure prophylaxis for this pandemic strain is reported to be effective even when administrated more than 48 h after exposure in household settings [18] , however, delays in administration are not specified. We cannot exclude the possibility that in our study, even delayed administration of oseltamivir prophylaxis may have prevented some people from becoming ill, although we anticipate the effectiveness of the intervention overall to be less in this setting than in households. Our study among 17 contact investigations showed an average total delay of 3,9 days between flight arrival and identification of contacts by passenger list, which is too late for effective PEP, and late for alerting on first symptoms of disease. Only in three contact investigations (18%), contact details were obtained within 48 h. However, after identification of passenger details, health authorities need time to actually trace the contact and administer PEP. It is highly unlikely that this was achieved within the same 48 h. We therefore conclude that contact investigation for provision of PEP as conducted here was ineffective. Regarding the awareness of ILI symptoms, Schiphol Airport handed all passengers on flights arriving from Mexico information leaflets on influenza A/H1N1 2009 with information on early symptoms and requesting them to seek medical advice in case of fever and respiratory symptoms such as coughing. Posters with this information were placed in passenger halls, to inform passengers arriving indirectly from Mexico via transit through other airports, or arriving from non-endemic areas with higher transmission (e.g. USA). As contact details were identified on average 3.9 days after exposure, however not contacted yet, we conclude that CT did not have additional value for timely achievement of increased awareness. It is not a new finding that contact tracing of flight passengers is a time-consuming procedure [8] . In one study among flight passengers during the pandemic in 2009, 52% (53/95) of the contacts were reached within 72 h [5] . In a measles contact investigation, 75% (202/ 275) of responding passengers were contacted within 72 h. In this study however, the diagnosis of measles was already suspected during the flight, and laboratory confirmation was initiated immediately after landing [19] . It also helped that many contacts were tourists staying at the same hotels, which facilitated tracing them. Our study shows that the longest delay before identification of contact details for an influenza index case is caused by the time between arrival and laboratory diagnosis (interval I, 2,6 days). This delay is a result of patients delay in seeking medical care, and doctor's delay, including laboratory confirmation. For influenza, the indicated laboratory test was Polymerase Chain Reaction, which takes several hours to obtain the result and in the beginning of the pandemic, the PCR test was not yet available in many laboratories. Patients delay was considerable however. It even took the seven passengers with date of onset before the flight, and therefore symptomatic during the flight, 1 to 2 days after arrival before laboratory confirmation was made. Also, none of the airline reported that these patients already were identified during the flight, nor that infection control measures were taken. For the indexes that became ill on the day of arrival, delay until laboratory confirmation still lasted 3 days (range 1-6 days). A prepandemic study by Sharangpani et al. among flight passengers showed that they are more willing to seek physicians care in case they developed flu-like symptoms when the perceived the pandemic as serious [20] . Leggat et al. demonstrated during the pandemic that only a minority (35,5%) of Australian citizens would cancel their air travel in case of cough and fever lasting more than 1 day. This was higher among persons who were more concerned about the pandemic [21] . In the Netherlands, the perceived severity of the disease decreased significant during this study period [22] . We expect that the delay until laboratory diagnoses in this study considerably is affected by patients delay seeking medical care, which might be better in diseases experienced as more threatening. Collecting passenger details from foreign airlines also caused considerable delay because of differences in time zones and the need to convince the concerned airline companies about the urgency to collect and hand-over passenger lists with contact details. Sometimes official request letters were necessary for legal reasons to release personal contact details. Dutch companies were easier to convince by Dutch health authorities to hand over passenger details. Our data show that contact details that were identified too late or not at all, indeed more often originated from non-Dutch than from Dutch airline companies. An internationally standardized contact tracing protocol, communicated with the International Civil Aviation Organization (ICAO) and International Air Transport Association (IATA), would facilitate the timeliness, and therefore effectiveness of contact tracing. Although one might expect differently, timeliness of CT for flights where PLC's were available, was not better than CT for flights without PLC. However, PLC's reduces the effort, in terms of staff support for airline companies and the municipal health service to collect useful passenger information considerably. PLC's were only used by Dutch airlines, who already were able to provide passenger lists relatively quickly. This also explains the limited attributed shortening in timeliness. Contact details on PLC's might be more accurate to trace the passenger than details provided by the passenger list or booking station. This is further investigated. This study has several limitations. As available data were recorded in days, and not in hours, it was not possible to determine the time intervals more precisely. As this was both with first and last date of the intervals, we expect no negative or positive bias. Secondly, the arrival date was used for date of exposure, while the actual exposure might have already taken place the day before at departure of the flight. This would imply an increase in delay and decrease the effectiveness of contact tracing. Also, we have no data if, and when contacts were actually reached and oseltamivir was administered. Since several steps were still required to reach the contacts after they were identified through passenger lists, this only would have lead to further delay in administrating prophylaxis. Further investigation into the timeliness of administration of prophylaxis among these contacts is initiated, to have insight in the delay of this last interval to facilitate future decisions on the effectiveness and necessity of contact tracing among flight passengers. Lastly, this study includes CT initiated at only one airport. CT procedures might be different at airports in other countries, which influences interval III. As this is not causing the main delay, we do not expect that in other countries CT would be much faster. We conclude that tracing close contacts among flight passengers during the initial phase of pandemic A/ H1N1 2009 was not effective, as timely provision of PEP could not be achieved in most cases. Most contacts came from an endemic area (Mexico) or areas with well known increased transmission during the first 2 months of the pandemic. The additional risk for those travelers of being a close contact during a long haul flight is small (3,5%) [5] . Furthermore, airline companies and/or Schiphol airport already provided contacts with information on the disease and its symptoms by. The benefit to inform them of the fact that they were contacts of a laboratory confirmed case did not justify the extra effort health authorities invested in contact tracing, especially during a period where public health officials, airports and airline companies were absorbed by efforts of other pandemic related control measures. In hindsight, the limited burden of disease of influenza A/H1N1 2009 did not justify contact tracing efforts. The main reason for flight contact tracing is raising alertness for possible exposure to uncommon infectious diseases, enabling early recognition and treatment of the disease and timely installation of control measures (e.g. SARS and viral hemorrhagic fevers). For some diseases, PEP is indicated as well. The risk assessment upon which the decision to install contact tracing is based should incorporate -apart from an evaluation of the severity and rarity of disease -an assessment of the required timeliness of effective control measures [23] . The expected time for laboratory confirmation of index cases and identification and tracing of contacts should be related to the maximum period during which quarantine, PEP or other control measures are effective in order to decide on the benefit of this time consuming procedure. Lastly, also cabin crew should be aware of their role of signaling infectious patients. In consultation with medical professionals, direct control measures can be installed, as well as medical evaluation after landing.

Validation of Self-swab for Virologic Confirmation of Influenza Virus Infections in a Community Setting

Few studies have investigated the validity of self-collected nose and throat swabs for influenza confirmation in community settings. We followed outpatients with confirmed influenza with sequential measurement of viral loads and applied log-linear regression models to the viral shedding patterns. Among 176 outpatients with confirmed influenza, the detection of virus and quantitative viral loads obtained from self-swabs was consistent with statistical predictions based on earlier and later measurements, suggesting that self-collected nose and throat swabs can be a valid alternative for virologic confirmation of influenza A or B infection in a community setting.

Few studies have investigated the validity of self-collected nose and throat swabs for influenza confirmation in community settings. We followed outpatients with confirmed influenza with sequential measurement of viral loads and applied loglinear regression models to the viral shedding patterns. Among 176 outpatients with confirmed influenza, the detection of virus and quantitative viral loads obtained from self-swabs was consistent with statistical predictions based on earlier and later measurements, suggesting that selfcollected nose and throat swabs can be a valid alternative for virologic confirmation of influenza A or B infection in a community setting. In community-based studies of acute respiratory illnesses, clinical specimens from the upper respiratory tract may be collected from patients at different stages of disease for virological testing. Although those clinical specimens are typically collected by trained healthcare professionals (HCPs) in a clinic setting, selfcollection by the patient at home may be a more acceptable, economical, and logistically feasible alternative. We investigated whether self-collected nose and throat swabs (NTSs) from patients in a community setting could provide a valid alternative for virologic confirmation of influenza A or B virus infection. Based on 2 similarly designed community-based studies, we modeled the viral shedding patterns from illness onset, adjusting for delays between clinical symptom onset and specimen collection, and compared the quantitative viral load measurements in self-swabs with model-based predictions. We conducted 2 separate community-based studies of influenza virus infection in Hong Kong with broadly similar protocols for recruitment and follow-up. In both studies, outpatients with recent-onset acute respiratory illness who presented within 48 hours of symptom onset were recruited; of individuals who provided informed consent, those with a positive result on a QuickVue Influenza A1B rapid diagnostic test (Quidel Corp) were invited to continue with follow-up. In one prospective, multicenter study (Influenza Resistance Information Study [IRIS] ), patients recruited between 20 January 2010 and 24 November 2010 were followed up to examine natural prevalence and/or emergence of resistance to antivirals among circulating influenza virus strains. In a separate household transmission study (HTS), eligible patients were recruited between 9 January 2008 and 29 September 2008 and followed up as part of a study investigating the effectiveness of nonpharmacological interventions [1] . In the IRIS, NTSs were collected by a trained HCP on days 1 and 6 after recruitment in the outpatient clinic and self-collected by subjects at home on day 3 after receiving detailed instruction on swab technique from the HCP at baseline. During the clinic visits, the nasal swab was collected by inserting and rotating a separate flocked sterile swab (Copan) through each nostril into the posterior nares, and throat swabs were collected by swabbing a sterile flocked swab on both the tonsillar fossae and posterior pharynx. The flocked end of the 2 nasal swabs and the throat swab were then transferred to a vial containing Copan Universal Transport Medium by breaking the prescored breaking point of the plastic swab shaft. Specimens collected in the clinic on day 1 and day 6 were stored directly in a clinic refrigerator at 4°C-8°C after collection. Face-to-face instruction on how to perform a nasal and throat swab on oneself or one's child was given to the patient or to the parents of children aged ,8 years by the HCP during the baseline visit on day 1. Patients also received a patient instruction leaflet and a kit containing the 3 swabs, an individually wrapped tongue depressor, a transport medium vial, and a sealable plastic bag. Patients kept the day 3 swabs in a refrigerator at home after collection and returned them to the clinic on the day 6 visit. All specimens were sent by courier in insulated transport container to the central laboratory at Erasmus Medical Centre within 7 days of collection. In the HTS, all NTSs were collected by a trained HCP at home visits on days 1, 4, and 7 after recruitment. Nasal swabs were collected by inserting and rotating a sterile plain swab (viscosetipped collection swab with a snappable plastic stick; EURO-TUBO) into the anterior nares, and throat swabs were collected by rubbing a second sterile swab against the tonsillar fossa. Both swabs were then snapped off into a tube containing viral transport medium (0.5% bovine serum albumin in Earle's balanced salt solution with antibiotic). Specimens were stored in an insulated transport container with at least 2 ice packs immediately after collection. Specimens were then either delivered directly or first stored overnight in a study outpatient clinic in a 2°C-8°C refrigerator and then delivered the next day to the central testing laboratory at Queen Mary Hospital by courier in ice boxes. Slightly different laboratory procedures were used in the 2 studies. For the IRIS, influenza A and influenza B matrix gene-specific reverse-transcription polymerase chain reactions (RT-PCRs) were performed as described elsewhere [2] . Dilutions of an electron microscopic-counted influenza virus A/PR/8/34 stock (Advanced Biosciences) and B/Lee/40 (Advanced Biotechnologies) were run in parallel for conversion of RT-PCR threshold cycle (Ct) values into a quantitative measurement of viral particles per milliliter (vp/mL) [2] . For the HTS, samples were eluted and cryopreserved at 270°C immediately after receipt in the laboratory. Specimens were then tested by a quantitative RT-PCR assay to detect the presence of influenza A or B virus and determine molecular viral loads in RNA copies per milliliter (copies/mL) using standard methods as described elsewhere [1, [3] [4] [5] . Previous studies have suggested that following influenza virus infection, viral load rises to a peak around the time of illness onset and then, for influenza A, declines approximately log-linearly over the subsequent 5210 days to undetectable levels and, for influenza B, plateaus with a more gradual decline [6, 7] . We specified multivariable linear regression models for the log viral load on the first and third measurement (typically 0 and 7 days, respectively, after recruitment), with the same slope but separate intercepts for each individual to allow for between-person variability in peak viral loads. We fitted separate models for each study and for influenza A and B and adjusted for age and oseltamivir treatment. Interaction terms with time were included to allow the slope of the regression line to vary by age and oseltamivir treatment. Viral loads for specimens with measured load below the lower limit of quantification (LLOQ) were imputed as half the LLOQ. This random-effects regression model constructed using the first and third measurement was used to predict viral loads expected on the second measurement (typically 3 days after recruitment), which were then compared with the observed viral loads on the second swabs, which were collected by the patients in IRIS and by an HCP in the HTS. We calculated mean differences with 95% confidence intervals (CIs) based on the t distribution. One hundred thirty-eight subjects with confirmed influenza A and 58 with confirmed influenza B were recruited in the IRIS in 2010, including 43% aged ,15 years (range, 2-85 years); 53% were female, and 53% were prescribed oseltamivir treatment. One hundred eighty-eight subjects with confirmed influenza A and 118 with confirmed influenza B were recruited into the HTS in 2008, of whom 73% were aged ,15 years (range, 0-79 years); 54% were female, and 25% were prescribed oseltamivir treatment. The demographic characteristics of subjects with influenza A versus B were similar. Oseltamivir treatment was more common during the period of peak pandemic A (H1N1) activity (data not shown). Among subjects with a positive RT-PCR result for influenza A at the first measurement and a self-swab available, 109 of 121 (90%) subjects in the IRIS had detectable virus in the selfcollected swab 2-5 days after illness onset. In the HTS with an HCP-collected swab, 132 of 183 (72%) subjects had detectable virus in the swab 2-7 days after illness onset. For influenza B, the corresponding statistics were 49 of 55 (89%) for the IRIS (2-5 days after onset) and 74 of 117 (63%) for the HTS (2-7 days after onset). Trends in influenza A viral load are shown in Figure 1A for 138 subjects from the IRIS and in Figure 1B for 188 subjects from the HTS. In the IRIS, the influenza A viral loads determined from swabs taken at the second measurement were slightly lower on average than the expected values based on the random-effects regression model ( Figure 1E ). The mean difference between observed and predicted viral load on the second measurement was 20.50 (95% CI, 2.69 to 2.31) log 10 vp/mL. In the HTS, viral loads determined from swabs taken at the second measurement were slightly higher on average than the expected values based on the random-effects regression model ( Figure 1F) , with a mean difference of 0.31 (95% CI, .08-.54) log 10 copies/mL. Trends in influenza B viral loads are shown in Figure 1C for 58 subjects from the IRIS and in Figure 1D for 118 subjects from the HTS. In both studies, the differences between observed and predicted viral loads on the second measurement were small and statistically insignificant ( Figure 1G and 1H) . In the IRIS, the mean difference was 0.16 (95% CI, 2.18-.51) log 10 vp/mL, and in the HTS, it was 0.14 (95% CI, 2.16-.43) log 10 copies/mL. Results from the HTS, in which all 3 swabs were collected by trained HCPs, showed that the viral load on the second measurement could accurately and reliably be predicted from a loglinear model based on the first and third measurements. Applying the same approach to the IRIS data, we found that viral loads from self-swabs on the second measurement were very similar to the viral loads that we would have expected if the second swab had been collected by a trained HCP. Our results therefore support the feasibility and validity of using self-swabs as an alternative approach to permit laboratory confirmation of influenza-associated illnesses in a community setting. Previous studies have demonstrated the feasibility of using parent-collected NTSs from children, in either a hospital setting or community setting, for laboratory confirmation of influenza and other respiratory virus infections without any significant loss in sensitivity [8] [9] [10] . Our results further extend this to selfcollected swabs by patients in the community setting, both for qualitative disease confirmation and quantitative viral load estimation. In the IRIS, 90% of self-collected swabs contained detectable influenza virus approximately 4-6 days after illness onset, indicating no substantial loss in sensitivity for qualitative virus detection through this approach. In the HTS, the lower proportion of specimens with detectable virus in the second swab can be attributed to the slightly longer average delay from illness onset. Our results also suggest that self-swabs work generally well for quantitative measurement of viral loads. For influenza B, there was no significant difference between those obtained from selfswabs and the values predicted from the other 2 swabs by HCPs ( Figure 1G and 1H) . For influenza A, overall trends in viral loads between the 2 studies also appear similar ( Figure 1A and 1B) . Although self-swabs from the IRIS were associated with a lower viral load than the predicted value ( Figure 1E and 1F), these results should be interpreted with some caution as differences in the type of swabs and transport media used, collection site and technique, delays between collection and transport to the laboratory, and laboratory procedures between studies may have led to artifactual differences. We therefore only compared the predicted and observed viral loads within each study because results from the 2 studies were not directly comparable. Although the log-linear model fit the data well and provided reasonable predictions of viral loads in the HTS (Figure 1 ), more complex models might better represent the decline in viral loads over time. Although we did not explicitly model the shedding patterns of different influenza A subtypes, we have not previously identified substantial differences [2, 5] . No reports in the literature exist on the validity of self-swab for longitudinal studies of influenza virus infection and illness in Figure 1 . A-D, Molecular viral loads on first and third measurements (circles ) and second measurement (crosses ) for influenza A from the Influenza Resistance Information Study (IRIS) (A ) and the household transmission study (HTS) (B ) and for influenza B from the IRIS (C ) and the HTS (D ). E-H, Difference between observed and expected molecular viral load (VL) at the second measurement, with a histogram summarizing the differences, for influenza A from the IRIS (E ) and the HTS (F ) and for influenza B from the IRIS (G ) and the HTS (H ). The second measurement (crosses ) was collected by self-swab in the IRIS (A, C, E, G ) and by a healthcare professional in the HTS (B, D, F, H ). For the IRIS, the lower limit of detection (LLOD) of the influenza A assay was 54 viral particles per milliliter (vp/mL), and the lower limit of quantification (LLOQ) was 131 vp/mL; the LLOD of the influenza B assay was 168 vp/mL, and the LLOQ was 194 vp/mL. For the HTS, the LLOD of the influenza A and B assays was 550 copies/mL, and the LLOQ was 900 copies/mL. a community setting. Because of the need for multiple sequential respiratory specimens over the course of illness, such studies typically require multiple clinic visits by the patients or multiple home visits by the HCPs and are thus costly and complicated, which may also affect study compliance. Self-swab would thus be an attractive alternative, and further validation of this approach would benefit the design of future community-based studies. Further studies employing collection of NTSs by both the patients and HCPs in a parallel or randomized fashion could help to enable finer calibration of the measurements obtained by self-swabs.

Molecular Mimicry as a Mechanism of Autoimmune Disease

A variety of mechanisms have been suggested as the means by which infections can initiate and/or exacerbate autoimmune diseases. One mechanism is molecular mimicry, where a foreign antigen shares sequence or structural similarities with self-antigens. Molecular mimicry has typically been characterized on an antibody or T cell level. However, structural relatedness between pathogen and self does not account for T cell activation in a number of autoimmune diseases. A proposed mechanism that could have been misinterpreted for molecular mimicry is the expression of dual T cell receptors (TCR) on a single T cell. These T cells have dual reactivity to both foreign and self-antigens leaving the host vulnerable to foreign insults capable of triggering an autoimmune response. In this review, we briefly discuss what is known about molecular mimicry followed by a discussion of the current understanding of dual TCRs. Finally, we discuss three mechanisms, including molecular mimicry, dual TCRs, and chimeric TCRs, by which dual reactivity of the T cell may play a role in autoimmune diseases.

Chronic autoimmune diseases are the by-product of the immune system recognizing self-antigens as foreign, which can lead to inflammation and destruction of specific tissues and organs (immunopathology) [1] . The impact of these diseases is global and heterogeneous with over 100 million people afflicted with more than 80 different autoimmune diseases [2] . While the etiology of autoimmune diseases is not fully elucidated, the causes are likely based on a combination of hereditary and environmental factors [3] . Although host genetic background contributes to the induction of an immune response to self, epidemiological and molecular evidence implicates infectious agents (viral and bacterial) as the principal environmental insults responsible for the induction of autoimmune diseases (reviewed in [4] [5] [6] ). Prolonged proinflammatory responses to infections have been associated with the initiation and exacerbation of autoimmune diseases (reviewed in [4, 7, 8] ). Inflammation is facilitated by proinflammatory cytokines such as type I interferon (IFN), interleukin (IL)-1β, IL-12, IFN-γ, IL-17, and tumor necrosis factor (TNF)-α (reviewed in [7, 9, 10] ). However, these proinflammatory cytokines are critical for clearance of pathogens, suggesting that environmental factors are able to divert the immune response towards immunopathogenesis. Although a number of immune cells are responsible for secreting proinflammatory cytokines, the primary cell types implicated in a vast majority of autoimmune disorders are autoreactive B and T cells, or antibody recognition of self [11] . Although a number of viruses and bacteria have been linked to the initiation of certain autoimmune diseases, identifying a particular virus or bacteria that is solely responsible for the induction of an autoimmune response is rare. This occurrence is due to the potential for multiple infections being involved in priming the immune system and other infections triggering disease, which could explain why no one viral infection has been conclusively linked to the development of immune-mediated autoimmune diseases [7] . However, there are a variety of examples of bacterial infections initiating and exacerbating autoimmune diseases. Streptococcus pyogenes is a gram-positive bacterium which causes group A streptococcal infection that is responsible for a number of diseases. The complications associated with S. pyogenes are rheumatic fever and glomerulonephritis. The infection causes the production of cross-reactive antibodies in response to the bacteria. Antibodies recognize the M protein (virulence factor) and the N-acetyl-β-Dglucosamine (GLcNAc) of S. pyogenes and cross-react with myosin leading to heart damage (reviewed in [8, 12, 13] ). Further evidence of molecular mimicry due to the production of cross-reactive antibody includes infection with gram-negative bacteria, such as Klebsiella pneumoniae and Campylobacter jejuni. Infection with K. pneumonia or C. jejuni leads to the production of cross-reactive antibodies able to recognize the self-antigens histocompatibility leukocyte antigen (HLA)-B27 and gangliosides, which induce ankylosing spondylitis and Guillain-Barré syndrome, respectively (reviewed in [8, 14] ). Examples of human autoimmune diseases with possible links with molecular mimicry are presented in Table 1 . The immune system has a number of mechanisms that are able to detect foreign pathogens by utilizing the major histocompatibility complex (MHC). This locus encodes the HLA genes and a variety of immune response (Ir) genes, thereby shaping the immune system that protects against pathogens. There are two main types of HLA antigens, HLA class I and class II. The function of HLA class I molecules is to present viral peptides at the surface of an infected cell to a T cell receptor (TCR) on a CD8 + T cell. The activation of these CD8 + T cells leads to the killing of the virally infected cell. This role of HLA class I, the identification of cells that are infected, explains why all nucleated cells have the capacity to express these MHC molecules. HLA class II molecules, in comparison, are expressed almost exclusively on the surface of dendritic cells, B lymphocytes, macrophages, endothelial cells, and activated T cells. Functionally, the HLA class II molecules present peptides to the TCR on CD4 + helper T cells. The engagement of the TCR by the peptide-MHC complex is necessary for the activation of CD4 + and CD8 + T cells, thereby leading to an effective adaptive immune response against an invading pathogen [15]. CD4 + T cells are central mediators of the adaptive immune response including cytokine secretion and cellular and humoral defenses against a pathogen. The HLA locus is extremely polymorphic leading to a heterogeneous population ensuring propagation of a species against novel pathogens. Unfortunately, this genetic heterogeneity adds to the complexity of identifying HLA genes implicated in autoimmune diseases. In addition to its role in protection against pathogens, a second critical role of the MHC and Ir genes is to safeguard against self-reactivity by restriction of the immune response to self. In this regard, the immune system has developmental checkpoints for the maturation of a T cell. As a naïve T cell expressing a pre-TCR migrates from the bone marrow to the thymus, rearrangement of α and β TCR genes occurs and T cells that have either too high avidity or lack of recognition of self-antigens are selected against and subsequently programmed for cell death. This selection mechanism for generating mature αβ TCRs is named central tolerance. Further, peripheral mechanisms of tolerance are able to suppress autoreactive T cells through certain subsets of cells including regulatory T cells (Tregs) that are able to inhibit self-reactive immune cells in the periphery. Unfortunately, there are a variety of mechanisms including molecular mimicry, bystander activation, exposure of cryptic antigens, and superantigens by which pathogens can aid in the expression of an autoimmune disease [16] [17] [18] [19] [20] [21] . Inflammation induced by exposure to a foreign antigen can lead to autoimmune diseases from cross-reactive epitopes (molecular mimicry). These epitopes are segments of foreign antigens which, when presented to either T or B cells in the context of the MHC, can activate CD4 + or CD8 + T cells. The induction of the immune response and subsequent proinflammatory cytokine release is critical for clearance of a virus or bacteria. However, a sustained proinflammatory response against specific host tissues can occur when there is sequence or structural homology between foreign antigens and selfantigens, termed molecular mimicry [18] . Although this concept has been associated with autoimmunity, there are instances where mimicry (cross-reactivity) provides protection for the host, termed heterologous immunity [22] . Cross-reactivity or mimicry between various strains of viruses or bacteria could help explain how protective immunity arises in certain individuals even in the absence of prior exposure to an emerging pathogen. This example of sequence homology in which molecular mimicry between viruses leads to protective immunity is in contrast to a pathogen mimicking host epitopes (reviewed in [11] ). Over 30 years ago, molecular mimicry by either a virus [18] or bacteria [23] was hypothesized to initiate and exacerbate an autoimmune response through sequence or structural similarities with self-antigens. Currently, molecular mimicry is the prevailing hypothesis as to how viral antigens initiate and maintain autoimmune responses which lead to specific tissue damage [18] . Initial work by Fujinami, Oldstone, and colleagues identified mouse antibodies to measles virus and herpes simplex virus (HSV-1) obtained from antibody-secreting B cell clones [18] . These antibodies were reactive to both intermediate filaments of normal cells and the proteins of measles virus and HSV-1, [114, 115] , reviewed in [116] thereby demonstrating a relatedness between host and viral antigens [18] . Further work by Fujinami and Oldstone used myelin basic protein (MBP), a nerve sheath protein containing an encephalitogenic T cell epitope in rabbits. The hepatitis B virus polymerase (HBVP) protein was found through computer analysis to share six consecutive amino acids with the encephalitogenic MBP epitope [16] , and when rabbits were sensitized with either MBP or HBV peptides, the rabbit's tissue serum reacted against MBP. Further, rabbits sensitized with the HBVP peptide developed central nervous system (CNS) pathology similar to rabbits sensitized with whole MBP protein or the MBP peptide [16] . Importantly, the rabbits sensitized with HBVP did not contract hepatitis but still developed encephalomyelitis and presented with a similar pathology as MBPsensitized mice. These experiments were the first experimental demonstration of molecular mimicry, whereby a microbial peptide with similar amino acid sequences to the self-peptide was able to activate autoreactive T cells and subsequently cause specific tissue damage. Immune cells of the adaptive immune response are specifically activated, but the hallmark of autoimmunity is the dysregulation of the immune system, especially T and B cells recognizing self-antigens as foreign. Activation of an autoimmune response could be enhanced by a variety of other, albeit, non-mutually exclusive non-specific mechanisms including bystander activation and superantigens. The difference between other non-specific mechanisms that initiate autoimmunity and molecular mimicry is that microbial mimics specifically direct the immune response towards a tissue and/or organ. Originally, T cell recognition was postulated to be highly specific and cross-reactivity was thought to be a rare phenomenon. However, the structural requirements for peptide binding by MHC class II molecules that are presented to T cells were found to be based on amino Linear sequence matches in amino acid motifs is not the only criteria for mimicry [32] . It has been hypothesized that self-reactive immune cells are primed by molecular mimicry and bystander activation, thereby sensitizing the immune cells and leading to a "fertile field" but no apparent disease. Subsequent environmental insults could induce these sensitized autoreactive cells to cause an autoimmune disease. Work from our laboratory demonstrated that recombinant viruses having molecular mimicry with self-CNS antigens were unable to initiate an autoimmune disease individually [38] . However, infected mice that were subsequently challenged, after viral clearance, with a non-specific immunologic insult developed disease [38] . Further, subsequent experiments showed that conventional inflammatory responses to specific pathogens were able to induce disease in animals primed with a molecular mimic to a CNS antigen [39]. Therefore, not only is the priming of the immune system necessary for an autoimmune disease but the milieu to which the primed immune cells are exposed is an important factor in initiating an autoimmune disease. Animal models of various autoimmune diseases have explored the role of molecular mimicry as a contributing factor ( Table 2) . The use of transgenic (tg) mice expressing virus proteins as transgenes in specific organs has been an important model for providing evidence for molecular mimicry. The expression of lymphocytic choriomeningitis virus (LCMV) viral antigens in pancreatic islet cells and the subsequent cross of this tg mouse with a TCR-tg mouse specific for LCMV glycoprotein resulted in an animal that only developed autoimmune disease if virally infected [40, 41] . These results demonstrated that "self"-reactive T cells are present in the periphery and the immune cells appear to remain quiescent until an appropriate signal (viral infection) triggers the T cells to respond. There are a variety of non-mutually exclusive factors that lead to a fully activated T cell, such as the quantity of peptide-MHC presented on the surface of antigenpresenting cells and TCR avidity. The interaction between the peptide-MHC and TCR is critical for the initiation of an adaptive immune response and clearance of a pathogen [15] . In order for T cells to reach maturity, the T cell goes through a number of developmental checkpoints leading to somatic recombination of various gene segments. The TCR αand β-chains are generated by V-D-J recombination, which leads to αβ TCRs expressed on the surface of T cells [42, 43] . Although it was believed that T cell signaling was mediated by a single antigen receptor, recent evidence demonstrates that T cells are capable of expressing functional dual Vα TCRs at a frequency of approximately 30% in humans and 15% in mice; however, an accurate number of dual specific TCRs is lacking due to the limited availability of anti-Vα monoclonal antibodies (mAbs) [44] [45] [46] . Interestingly, in contrast to the high frequency of dual expressing Vα T cells, only 1% of humans and 5-7% of mice express two β-chains due to allelic exclusion mechanisms, but the frequencies of dual Vβ TCRs have been found to be higher with age and in TCR-tg mice [47] [48] [49] . Expression of multiple TCR Vαs on the surface of a T cell is the result of simultaneous rearrangement of both TCRα loci during thymocyte development [50-52]. Further, TCR Vβ-chains preferentially bind to certain Vαchains leading to differential expression of chimeric TCRs on the surface of T cells [51, 53, 54]. Due to the heterogeneity of TCRs normally expressed in the periphery of humans and mice, TCR-tg mice have been used to track and determine the fate of T cells expressing dual TCRs. The use of TCR-tg mice has led to the identification of a potential role for dual TCRs in a variety of conditions including graft-versus-host disease, human immunodeficiency virus infection, inflammatory bowel disease, T cell leukemia, T cell lymphoma, and MS [55-61]. The expression of dual TCRs by the same T cell has been proposed to be a potential mechanism for autoimmune disease. Normally, high avidity self-reactive T cells are thymically depleted, but it has been hypothesized that the expression of a self-TCR on a T cell is lower when presented in the context of a second TCR, thereby providing a cover for high avidity self-TCRs from both central and peripheral tolerance. Blichfeldt et al. [62] demonstrated that dual tg-TCRs, which have lower expression of each TCR on the surface of a T cell, needed higher concentrations of peptide, presented by MHC, to induce a similar T cell proliferative response compared to a single receptor T cell. A potential role of dual TCRs in autoimmunity is in the rescue of autoreactive T cells from thymic selection. For example, the double tg mouse for autoimmune diabetes, in which the mice express a TCR specific for peptide 111-119 of hemagglutinin (HA) (TCR-HA) under the control of the rat insulin promoter and develop spontaneous diabetes and insulitis [63], were used to determine how T cells could escape tolerance mechanisms even if the antigen was ubiquitously expressed [64]. Low expressing TCR-HA coexpressing T cells were more effective at transferring diabetes than TCR-HA high dual TCRs, suggesting that the surface level expression of a dual TCR can be modulated by a second TCR expressed on the same T cell, thus "escape" of autoreactive T cells could be the first step in an autoimmune disease. The "trigger" of an autoimmune disease could be linked to environmental insults, such as viruses. A T cell co-expressing TCRs specific for a self-antigen and a foreign antigen could potentially allow for autoreactive T cells to be activated if the host is exposed to that foreign antigen. The activation of a subset of T cells could than lead to tolerance being broken and the initiation of an autoimmune disease if these T cells experienced a particular organ or tissue that expressed the self-antigen for the other TCR expressed at the surface of the T cell. In support of a role for dual TCRs in autoimmune diseases, work performed in our laboratory characterized autoreactive CD8 + T cells isolated from the spleens of Theiler's murine encephalomyelitis virus (TMEV)infected SJL/J mice [65] . In vitro assays testing CD8 + T cell killing activity found a population of CD8 + T cells that killed uninfected syngeneic cells [65] . Adoptively transferring these TMEV-specific autoreactive CD8 + T cells into non-infected SJL/J mice caused CNS pathology [65] . Further support for the importance of the mechanism by which viral infection could induce an autoimmune disease through dual TCR-expressing T cells was performed by Ji et al. [61] using MBP(79-87) TCR-tg mice [66] . Cytometric phenotyping, in vitro CD8 + T cell killing assays, and adoptive transfer experiments were used to track the expansion and killing capacity of Vα8Vβ8 MBP (79-87)-specific TCR and Vα8Vβ6-vaccinia virusspecific TCR. Infection of these tg mice with vaccinia virus induced autoimmune disease, thus demonstrating a virus triggering an autoimmune disease through dual TCR expressing T cells [61] . Although several tg TCR β-chains have been described on peripheral T cell [61, 67-70], there is no evidence that co-expression of dual TCRs leads to autoimmunity without the use of TCR-tg mice. As described above, current work in our laboratory has characterized TMEV-specific autoreactive CD8 + T cell clones derived from a wild-type animal, and these autoreactive TMEV-specific T cell clones express dual TCRs (manuscript in preparation). Importantly, we were able to induce CNS pathology in naïve SJL/J mice by adoptively transferring the TMEV-specific clones. Although further work is needed in order to identify the self-antigen that activates these CD8 + T cells, to our knowledge these results are the first demonstration of an autoimmune disease initiated by a dual expressing TCR characterized in the virus' natural host. Taken together, three possible mechanisms could explain how the dual reactivity of the TCR may play a role in autoimmune diseases (manuscript in preparation). The first mechanism is molecular mimicry, whereby the induction of an autoimmune response to self is due to a single TCR recognizing both a virus and a self-antigen. The second mechanism is the expression of dual TCRs on a single T cell, where one TCR is able to recognize a microbial antigen and the other TCR recognizes self. The third mechanism involves a T cell expressing chimeric TCRs generated from either a single Vα combining with two different Vβs or a single Vβ combining with two different Vαs, resulting in a T cell with the potential of expressing two different chimeric TCRs specific for a self-antigen and a foreign antigen.

Low usage of government healthcare facilities for acute respiratory infections in guatemala: implications for influenza surveillance

BACKGROUND: Sentinel surveillance for severe acute respiratory infections in hospitals and influenza-like illness in ambulatory clinics is recommended to assist in global pandemic influenza preparedness. Healthcare utilization patterns will affect the generalizability of data from sentinel sites and the potential to use them to estimate burden of disease. The objective of this study was to measure healthcare utilization patterns in Guatemala to inform the establishment of a sentinel surveillance system for influenza and other respiratory infections, and allow estimation of disease burden. METHODS: We used a stratified, two-stage cluster survey sample to select 1200 households from the Department of Santa Rosa. Trained interviewers screened household residents for self-reported pneumonia in the last year and influenza-like illness (ILI) in the last month and asked about healthcare utilization for each illness episode. RESULTS: We surveyed 1131 (94%) households and 5449 residents between October and December 2006 and identified 323 (6%) cases of pneumonia and 628 (13%) cases of ILI. Treatment for pneumonia outside the home was sought by 92% of the children <5 years old and 73% of the persons aged five years and older. For both children <5 years old (53%) and persons aged five years and older (31%) who reported pneumonia, private clinics were the most frequently reported source of care. For ILI, treatment was sought outside the home by 81% of children <5 years old and 65% of persons aged five years and older. Government ambulatory clinics were the most frequently sought source of care for ILI both for children <5 years old (41%) and persons aged five years and older (36%). CONCLUSIONS: Sentinel surveillance for influenza and other respiratory infections based in government health facilities in Guatemala will significantly underestimate the burden of disease. Adjustment for healthcare utilization practices will permit more accurate estimation of the incidence of influenza and other respiratory pathogens in the community.

As the 2009 influenza A (H1N1) pandemic highlighted, surveillance for influenza is now a worldwide priority. [1, 2] At the 58 th World Assembly in 2005, The World Health Organization adopted a resolution calling for Member States to fortify and coordinate national strategies to prepare for an influenza pandemic, including establishment of surveillance systems for human influenza. [3] To assist with the development of standardized influenza surveillance systems in the Americas, the Pan American Health Organization (PAHO) and the United States Centers for Disease Control and Prevention (CDC) developed a generic protocol for influenza surveillance incorporating two sentinel surveillance systems, one hospital-based system for severe acute respiratory infections (SARI) and SARI-related mortality and another for influenza-like illness (ILI) based in ambulatory clinics. [4] Sentinel surveillance for influenza can provide information on trends in viral circulation patterns and seasonality, along with virus characteristics to help guide decisions on vaccine composition. However, healthcare seeking behaviors can affect who accesses care at the sentinel site, limiting the ability to gather information to guide public health policies. Without understanding patterns of healthcare seeking behavior, it is not possible to calculate the burden of disease, generalize findings to a larger population or identify risk groups. Healthcare utilization surveys (HUS), one method of determining the healthcare utilization practices for specific diseases in defined populations, have been conducted in several countries. [5] [6] [7] [8] [9] [10] [11] [12] In HUS, random samples of the catchment population are interviewed with respect to their healthcare seeking and treatment behaviors during recent episodes of disease. These data can be used in a number of ways to support the interpretation of information from sentinel surveillance sites: first, to establish correction factors for estimates of incidence in the community based on numbers of cases presenting at the sentinel surveillance site, including the incidence in particular population sub-groups; second, to describe the actual catchment population accessing healthcare at the sentinel site to determine generalizability; and third, to identify other healthcare providers who may be recruited to participate in the surveillance system. To inform the establishment of a surveillance system for influenza and other respiratory infections, and the implementation of the PAHO/CDC standard protocol for influenza surveillance in Guatemala, we conducted a HUS among residents of the Department of Santa Rosa, Guatemala to describe the healthcare seeking behavior for acute respiratory illnesses. Guatemala, with a population of 12,755,366 in 2008, had a gross national income per capita of $2680 and is considered a middle-income country by the World Bank (http://data.worldbank.org/indicator/NY.GNP.PCAP.CD, accessed on 1 September 2010). Guatemala is divided into 22 departments, which are further subdivided into 10-29 municipios (similar to counties), made up of multiple communities. The Guatemalan Ministry of Public Health and Social Welfare (MSPAS) provides free healthcare in several different settings, including hospitals, health centers, health posts, and outreach centers. Hospitals and health centers are staffed by physicians and nurses, whereas health posts are staffed by nurses. Outreach centers provide preventive and primary healthcare but are only visited by trained medical staff a few days each month. In addition to the MSPAS facilities, formally-employed workers who contribute to the Guatemalan Institute of Social Security (IGSS) can receive healthcare from IGSS hospitals and health centers, which are concentrated in Guatemala City. Other non-governmental locations where people may seek healthcare are private hospitals and clinics, pharmacies, drug shops, and traditional healers and midwives. Communities in Guatemala vary in their access to healthcare depending on their size and location (e.g., urban vs. rural). Santa Rosa is a mostly agrarian department located in the southeastern part of the country approximately 80 km from Guatemala City. The population in 2006 was 308,522 residing in 14 municipios with an estimated 768 communities. Cuilapa is the department's capital city. In contrast to the country as a whole, which is almost half Amerindian indigenous, only 3% of Santa Rosa's residents are Mayan or Xinca, and Spanish is spoken by approximately 91% of the inhabitants. The mortality rate of children <5 years of age for Santa Rosa is 58 per 1000 live births, significantly higher than the average for the country (45 per 1000 live births). [13] Government-run healthcare facilities within the department include one hospital (the National Hospital of Cuilapa, 176 beds), 14 health centers (one in each municipio) and 56 health posts in the outlying communities. There is one IGGS facility that treats only patients involved in motor vehicle accidents. Two small private hospitals as well as approximately 133 private ambulatory clinics are available for those who choose to pay for healthcare. Additionally, healthcare services can be sought from more than 114 pharmacies or drug shops, and an unknown number of traditional healers, midwives and community healthcare workers. Healthcare may also be accessed in Guatemala City and neighboring departments. We conducted a cross-sectional survey to determine healthcare utilization patterns for acute respiratory, diarrhea, neurologic and febrile illnesses: we report here only the results for acute respiratory infections. We used a stratified, two-stage cluster sampling procedure. Communities in the 2002 Guatemalan census were stratified as to whether or not they had a hospital or health center located in their community. As the first stage of sampling, 30 communities were selected within each stratum, using probability proportional to the population of each community, for a total of 60 clusters. Maps detailing household locations were obtained from the Guatemalan Census Bureau for these communities and 20 houses were randomly selected for a total of 1200 houses. Interviews were conducted in person from October 1 through December 13, 2006. All persons who had lived in the house for at least six of the preceding 12 months were considered members of the household and eligible for inclusion, including persons deceased at the time of the survey if they had been resident during the reference period. Infants <6 months of age were included if they had lived in the household since birth. Households were excluded if a head of household or consenting adult was unavailable after visiting the house on three separate occasions over at least two days, or if the household head declined to participate. Excluded households were not replaced. If a house was abandoned or no longer existed, another house was randomly chosen for inclusion. The adult respondents from each household were read a consent statement and asked to give verbal consent for their household's participation. The protocol for this study was reviewed and approved by the institutional review boards of the Centers for Disease Control and Prevention (Atlanta, GA) and the Universidad del Valle de Guatemala (Guatemala City, Guatemala) and approved by the MSPAS (Guatemala City, Guatemala). The sample size of 600 households per strata (1200 total) was based on calculations for diarrhea, a more common syndrome, rather than for pneumonia as information was not available on the expected incidence of pneumonia. However, assuming a 10% household nonresponse rate, and 4.8 persons per household, a sample of 600 households per stratum should yield between 39 and 156 persons with pneumonia, assuming an annual incidence of between 1.5% and 6.0%, respectively. Given a design effect of two due to the clustering of pneumonia cases by community and household, this sample would be large enough to estimate the proportion of the population seeking healthcare outside of the house for pneumonia with a precision of 10%, assuming 70% of persons with pneumonia seek care for their illness. A structured survey was completed for each participating household with information on both household-and individual-level characteristics. All household members were enumerated and an adult proxy was interviewed for children <15 years old or older residents not present at the time of the interview to determine whether any household member met any of the case definitions (mild or severe respiratory, diarrhea, acute febrile and acute neurologic illness) during the prescribed time period. A clinical history was obtained for each illness episode (if more than two episodes of the same illness were reported, the most recent illness episode was used as the reference) along with a history of healthcare treatment seeking. Proxies of household residents who died but met the case definition for one of the illnesses in the relevant time period before death were also administered the illness-specific forms. A case of severe respiratory illness, referred to as pneumonia, was defined as self-reported cough and difficulty breathing for two or more days, or a physician-diagnosis of pneumonia; the reference period was the 12 months prior to the interview. This case definition has been used in several studies of self-reported pneumonia in the community [12, 14] and is based on questions that were moderately sensitive and specific for pneumonia from a World Health Organization verbal autopsy questionnaire. [15] Additionally, severe pneumonia was defined for children <3 years old who met the pneumonia case definition as any of the following: blue lips and/ or nails, inability to breastfeed or drink, convulsions, unconsciousness or decreased activity. For those ≥3 years old who met the pneumonia case definition, severe pneumonia was defined as fast breathing with confusion. A case of mild respiratory illness, referred to as ILI, was defined as subjective fever with either cough or sore throat in the 30 days prior to the interview. If a respondent reported both an ILI and pneumonia for the same month, the ILI data were excluded. Outpatient providers were defined as all sources of healthcare that did not admit patients for overnight stays, and included government and private ambulatory clinics, pharmacies, drug shops, the IGSS and traditional healers. Inpatient providers included government and private hospitals. Survey forms were received at the offices of the CDC-UVG Collaboration at the Universidad del Valle de Guatemala for optical scanning into a database using the Cardiff Teleform system (Vista, CA). Each Teleform entry was checked manually with the original forms to ensure accurate scanning and coding. Socioeconomic status (SES) was estimated using a wealth index generated using the factor effects derived from the first principle component of a principle component analysis of household goods, house construction material, source of water supply, source of cooking fuel and sanitation facility; the wealth index was categorized into quintiles with households weighted by number of residents and sample weights. [16, 17] To account for the complex survey design, sample weights were applied in all analyses. We used the Wald Chi-square statistic to test for differences between proportions and logistic regression to test for trends related to age and SES. Analyses were conducted with SAS version 9.1 (SAS Institute, Cary, NC) using PROC SURVEYFREQ or PROC SURVEYLOGISTIC. We approached 1200 households but residents could not be reached at 33 (3%) locations after three visits, and in 36 (3%), the household head declined to participate. We interviewed residents from 1131 (94%) households and gathered information on a total of 5449 persons of which 2806 (52%) were female and 586 (12%) were children <5 years old (Table 1) . We found 323 persons (6%, 95% confidence interval [CI] 6-7%) who met the pneumonia case definition in the previous year. Almost all (87%) met the case definition with self-reported cough and difficulty breathing for at least two days; 2% reported only a physician's diagnosis of pneumonia; and 12% reported both. There were 60 cases (11%, 95% CI 9-13%) of pneumonia reported among children <5 years old, and 263 cases (6%, 95% CI 5-6%) among persons aged five years or older. Among the children <5 years old, 31 (6%, 95% CI 5-7%) met the case definition for severe pneumonia. The proportion of pneumonia cases reported by month increased from October 2005 through September 2006 (Figure 1 .) More than half of the pneumonia cases were reported from the last five months prior to the survey. There were 628 (13%, 95% CI 12-14%) persons who reported ILI in the previous month. A case of ILI was reported by 106 (19%, 95% CI 17-20%) children <5 years old and 522 (12%, 95 CI 11-13%) persons aged five years or older. The most common symptoms reported by persons with pneumonia were difficult breathing (100%), cough (99%), and feverishness (88%) ( Table 2 ). The mean duration of illness of all persons with pneumonia was 13 days (range 2-120); more than one-quarter had symptoms for seven days or more. Among persons who reported ILI, the most common symptom besides feverishness (100%) was sore throat (97%), headache (89%) and cough (88%). Signs of lower respiratory tract infection, such as difficult or fast breathing and wheezing, were less common among person who reported ILI than those with pneumonia. The mean duration of illness among persons with ILI was seven days (range 1-90); more than half of person who reported an ILI had symptoms for seven days or more. The age distribution of persons with pneumonia was significantly different from the surveyed population without pneumonia (P<0.0001), with more children <5 years and adults ≥60 years old among the persons reporting pneumonia than among the surveyed population without pneumonia (Table 1) . Similarly, the age distribution of those with ILI was younger than the surveyed population without ILI (P = 0.001). The distribution of the person who reported ILI by household wealth index was significantly different from those without ILI (P<0.0001) with more cases among persons in the lowest wealth category and fewer in the wealthiest category. There was no difference in household wealth between persons with and without pneumonia (P = 0.14). Among the 60 children <5 years old reporting pneumonia in the last year, 55 (92%) sought care outside the home. All subsequent analyses of healthcare-seeking behavior are based on those who sought care outside the home. Sixteen (27%) children sought care from more than one source. Hospitals were consulted by 17 (25%) children <5 years old with pneumonia, and most were government hospitals (Table 3) . Nine (12%) children <5 years old with pneumonia were admitted for at least one night in a hospital. Outpatient care providers were visited by 38 (75%) children <5 years old with reported pneumonia. Overall, the most frequently reported source of healthcare for children <5 years old with pneumonia were private ambulatory clinics, which attended to more than half the reported cases. More than half (55%) of the children <5 years old with reported pneumonia received care at least once during their illness from government facilities, either government hospitals or ambulatory clinics. Among the 263 persons five years or older who reported pneumonia in the last year, 199 (73%) sought healthcare outside their home, with 21 (8%) seeking care from more than one source. Among persons in this age group who sought care outside the home, 28 (12%) sought care at hospitals (Table 3) , and 8 (4%) were admitted for at least one night. Government hospitals provided most of the hospitalized care. Among outpatient care providers, the most frequently sought source of care were private clinics, which provided care to 65 (31%) persons with pneumonia aged five years or older, along with government ambulatory clinics (46, 27%). A considerable proportion (16%) of persons with pneumonia aged five years or older sought care at pharmacies. Care for ILI was sought outside the home by 87 (81%) children <5 years old, and 6 (6%) sought care from multiple sources (Table 3) . Government clinics were the source of healthcare most often consulted by children <5 years old for ILI; 34 (41%) children <5 years old reported seeking care at a government clinic, whereas 20 (20%) reported consulting a private clinic. Hospitals were consulted by 6 (5%) children <5 years old for ILI, and 4 (5%) were hospitalized for one night or more. Care was sought at pharmacies and drug shops for nearly one-third of children <5 years old with ILI. Care for ILI was sought outside the home by 337 (65%) persons aged five years or older, and 13 (3%) consulted multiple sources. Government clinics were consulted for ILI by 111 (36%) persons aged five years or older. Pharmacies were consulted for ILI by 110 (29%) persons aged five years or older. One (0.1%) ILI patient five years or older was hospitalized for more than one night. Among the respondents with pneumonia who did not seek healthcare for their illness, the perception that their illness was not severe enough to warrant treatment (28/69, 42%) and the cost of treatment (13/69, 20%) were the major reasons cited for not seeking care. Among persons with ILI who did not seek healthcare for their illness, insufficient severity of illness (68/204, 31%), cost of treatment (37, 18%), lack of medical services (13, 9%) and spontaneous improvement (19, 7%) were the major reasons cited. Sociodemographic and illness characteristics associated with seeking treatment at government facilities There was no significant association between sex of the respondent and whether care was sought for pneumonia (P = 0.34) or ILI (P = 0.15) at a government hospital or clinic (Table 4 ). Children <5 years old were more likely to receive healthcare for pneumonia and ILI at government facilities than persons aged five years or older, but this difference was statistically significant only for Numbers will not necessarily add up to 100% because more than one healthcare provider can be consulted in the course of an illness. Percentages are calculated using sample weights. pneumonia (P = 0.03). There was a significant inverse trend across SES status with persons of higher socioeconomic status less likely to seek care for pneumonia and ILI at government facilities (P = 0.005 and P = 0.001, respectively). The duration of illness was not associated with consultation at a government facility for either pneumonia (P = 0.37) or ILI (P = 0.25). Severity of pneumonia was not associated with seeking care from a government facility (P = 0.13). We conducted a HUS to help inform establishment of a sentinel surveillance system for pneumonia and influenza-like illness in Santa Rosa, Guatemala, and found that private clinics are the single most important source of healthcare for pneumonia both in children <5 years old and older persons. For more mild ILI, both children <5 years old and persons aged five years or older are more likely to consult government clinics than other sources of health care. These findings suggest that in Santa Rosa, sentinel surveillance for pneumonia in government hospitals will significantly underestimate the burden of disease, by up to 75% for children <5 years old and 88% for persons aged five years and older. Government healthcare clinics will underestimate the number of cases of ILI by about 59% for children <5 years old and 64% for persons five years and older. Our study is consistent with other studies of healthcare-seeking behavior in Guatemala and Central America that have found private clinics to be common sources of healthcare. Van der Stufyt et al. reported that more than 40% of Guatemalan families sought healthcare for their children <5 years old from private physicians, compared with 26% consulting a governmental health center. [18] Focus group interviews from three countries in Central America found that persons considered the healthcare obtained through private physicians and clinics preferable to public options because of prompt attention and a perception that healthcare is better. [19] Another study evaluating healthcare utilization among children in rural Guatemala who reported a diarrheal or respiratory illness found private physicians were more likely to be consulted by households with higher income. [20] The population that uses government health clinics and hospitals for respiratory illnesses in Santa Rosa is younger and poorer than the general population. We found a trend for decreasing use of government facilities with increasing age and household wealth. These findings should be taken into account to improve the generalizability of burden of disease estimates made from sentinel surveillance data. The results of this study are subject to several important limitations. We used a case definition that required self-report or report by a proxy of an illness that occurred up to one year prior to interview. It is well known that such self-reports are limited by recall decay, and recent episodes are more likely to be recalled than earlier episodes. [21] This could be noted in our data that demonstrated a decreasing report of pneumonia with increasing time before the survey. As long as recent illness episodes do not differ from prior episodes with regard to patterns of healthcare-seeking behaviors, there is no reason to believe that recall decay causes bias with regard to these variables. Because our case definition is based on self-report, there is substantial potential for misclassification, especially between mild and severe acute respiratory illness, and this can be seen in the report of some lower respiratory tract symptoms among respondents with ILI. However, the behaviors associated with episodes of ILI compared to pneumonia (lower probability of seeking care outside the home, less likely to seek treatment at a hospital) are suggestive of a more mild illness and we are reasonably confident that we have described the healthcare seeking behaviors that are broadly associated with both pneumonia and ILI. An additional limitation is our sample size, which is too small to permit age to be stratified into more than two groups, restricting our ability to model healthcare-seeking behaviors more precisely by smaller age groups. Finally, it is not clear whether results from one area of Guatemala with a significantly lower indigenous population than the rest of the country can be generalized to the nation as a whole. Despite the limitations of this and similar surveys, our findings indicate that Guatemala and other countries in the region can improve the estimation of the burden of influenza and other respiratory pathogens from sentinel surveillance by taking healthcare utilization into account. As a large proportion of the population with respiratory disease in Guatemala does not attend government health facilities for treatment, this approach could help correct government surveillance data for missing cases and facilitate comparison of the burden of influenza with other countries.

Understanding the clinical spectrum of complicated Plasmodium vivax malaria: a systematic review on the contributions of the Brazilian literature

The resurgence of the malaria eradication agenda and the increasing number of severe manifestation reports has contributed to a renewed interested in the Plasmodium vivax infection. It is the most geographically widespread parasite causing human malaria, with around 2.85 billion people living under risk of infection. The Brazilian Amazon region reports more than 50% of the malaria cases in Latin America and since 1990 there is a marked predominance of this species, responsible for 85% of cases in 2009. However, only a few complicated cases of P. vivax have been reported from this region. A systematic review of the Brazilian indexed and non-indexed literature on complicated cases of vivax malaria was performed including published articles, masters' dissertations, doctoral theses and national congresses' abstracts. The following information was retrieved: patient characteristics (demographic, presence of co-morbidities and, whenever possible, associated genetic disorders); description of each major clinical manifestation. As a result, 27 articles, 28 abstracts from scientific events' annals and 13 theses/dissertations were found, only after 1987. Most of the reported information was described in small case series and case reports of patients from all the Amazonian states, and also in travellers from Brazilian non-endemic areas. The more relevant clinical complications were anaemia, thrombocytopaenia, jaundice and acute respiratory distress syndrome, present in all age groups, in addition to other more rare clinical pictures. Complications in pregnant women were also reported. Acute and chronic co-morbidities were frequent, however death was occasional. Clinical atypical cases of malaria are more frequent than published in the indexed literature, probably due to a publication bias. In the Brazilian Amazon (considered to be a low to moderate intensity area of transmission), clinical data are in accordance with the recent findings of severity described in diverse P. vivax endemic areas (especially anaemia in Southeast Asia), however in this region both children and adults are affected. Finally, gaps of knowledge and areas for future research are opportunely pointed out.

Plasmodium vivax is the most geographically widespread species of Plasmodium causing human disease, with most cases reported in Central and Southeast Asia, in the horn of Africa and in Latin America [1] . It is considered to be a potential cause of morbidity and mortality amongst the 2.85 billion people living at risk of infection, excluding the large African populations who are mostly Duffy negative and, therefore, naturally less susceptible to this infection. However recent data suggest that the parasite is evolving and may use alternative receptors other than Duffy (DARC) for erythrocyte invasion [2] . It is estimated that 5.5% of the population under risk live in the Americas [3] . The major biological characteristic of this parasite is the presence of liver hypnozoites responsible for the frequent relapses, which add a substantial number of cases to the general burden of the disease, what is being faced as one of the most challenging bottlenecks for vivax malaria eradication [4] . Although often regarded as causing a benign infection, there is recent increasing evidence that the overall burden, economic impact, and severity of P. vivax have been underestimated, in part due to a bias in the scientific literature which traditionally devoted most of its attention to the more lethal parasite Plasmodium falciparum, probably as a reflection of a more substantial funding [5] . Until 16 October 2011, the search in MED-LINE using P. vivax as keyword retrieved 5,026 indexed abstracts; using P. falciparum, on the other hand, retrieved almost five times more abstracts: 25, 807 . Even in places where P. vivax represents the major local problem to be tackled, clinical research is still focused on P. falciparum [6] . There is robust evidence in the past decade from hospital-based studies in India and Indonesia that P. vivax is able to cause severe disease [7, 8] . Some authors argue that this clinical severity may only now be properly recognized and announced by researchers in the field, but these complications apparently are not new from a historical perspective [9] . Actually, the case fatality rate (CFR) related to malarial infections in the English marshes during the 16th and 17th centuries, corresponding to the Little Ice Age, suggest that P. vivax (a parasite more prone to persist in vectors even under low temperatures) may have killed part of this population already victimized by famine [10] . During the first half of the 20th century, malariotherapy in patients with neurosyphilis, using essentially the 'nonsevere' P. vivax parasite, led to diverse complications, CFR ranging from 3.3 to 30.3% [11] . The major related complications in these co-infected patients were liver damage, ruptured spleen, jaundice, delirium, uncontrolled vomiting and persistent headaches [11] . That reinforces the concept that P. vivax infection may synergize with other co-morbidities resulting in more complicated disease. Added to local geographical and social determinants, wide Annual Parasite Incidence (API) and CFR variations due to this species are seen around the world. In summary, P. vivax, which has long been neglected and mistakenly considered 'benign' [12] , is receiving an increasing amount of importance in the debates taking place on malaria epidemiology and control, drug resistance, pathogenesis and vaccines [13] . As reviewed elsewhere, the good clinical characterization of severe disease in vivax infection is the first step to understand how the inflammatory response to this parasite contributes to pathogenesis [14] . Traditionally, Brazil has been responsible for almost half of all cases of malaria in Latin America. In 2009, 308,498 cases of malaria were reported in this country (257,571 caused by P. vivax), representing 54.9% of all the malaria reported in the Americas [15] . Cases are virtually restricted to the Amazon Basin (constituted by the states of Amazonas, Acre, Roraima, Amapá, Pará, Tocantins, Rondônia, and parts of Mato Grosso and Maranhão). Amazonian urban agglomerations under continuous economical development trigger intense migration flows, such as in the city of Manaus (in the Western Brazilian Amazon), helping to maintain the disease under endemic levels [16, 17] . Malaria in Brazil is mostly related to P. vivax since the 1990s, when the available tools for control at the moment were put together and intensified, such as the fast diagnosis through thick blood smear (TBS) in all febrile patients, and free access to anti-malarials, integrated through a decentralized primary care-centred public health system [18] . Allied to that, an active community of local malariologists has been persistently identifying the profile of anti-malarial resistance with permanent counseling to the Brazilian Ministry of Health, which responds promptly to these evidences, changing the first line regimens [19] . As the sexual forms (which are infective for the vector) of P. falciparum generally appear later in the course of infection, opportune diagnosis and treatment tend to have a high impact on reducing the transmission intensity of this species but the same is not true for P. vivax, whose gametocytes are present in the very first days of the infection, before efficacious treatment is usually started. In 2008, 59% of all malaria cases registered in the Brazilian Amazon were treated in the first 48 h after appearance of symptoms (SIVEP-Malaria, 2009). These public health measures allied to a regularly updated online information system also impacted the number of deaths related to P. falciparum, which were not more than 58 in 2009 (Brazilian Ministry of Health, 2010). As a consequence, even in the non-indexed literature, severity due to P. falciparum is not frequently reported anymore in Brazil. Brazil has reported 85% of its cases related to P. vivax in 2010, which puts this country in a peculiar epidemiological situation, as one of the few countries around the world with P. vivax predominance. The impact of P. vivax/P. falciparum co-infections or simultaneous circulation of both species with similar frequencies in a given population, upon the immunological status and clinical presentation of malaria is still unclear [20, 21] , but most probably clinical data from population from certain areas should not be extrapolated to other areas in distinct epidemiological conditions. Actually, the lack of data on clinical presentation of P. vivax infection allied to the several particularities of this region, including the diverse genetic background of its population, implicate that the generalization of the findings from Southeast Asia may be inappropriate. In Brazil, in 1903, the young physician Carlos Chagas (most known for the discovery of American trypanosomiasis afterwards) wrote his MD thesis on the haematological complications of malaria, which, at that moment, also occurred in the non-Amazon area. His major findings in studying P. vivax patients were severe anaemia, splenomegaly, leukopenia, cachexia and jaundice associated to concomitant staphylococcal disease [22] . Bone marrows were also analysed in these patients with no conclusive findings. Later on, during the 1940s, Djalma Batista in Manaus described a series of malarial cases from his outpatient clinics in whom large splenomegaly, cachexia and minor bleeding were frequent among those with the 'benign' tertian malaria [23] . More recently, from 1998 to 2008, 234 deaths related to vivax disease were officially reported to the Brazilian Ministry of Health [18] , and an increase in the hospitalization trends for vivax patients was published in a tertiary care hospital from Manaus [24] . To complicate matters, these facts parallel a lack of robust biomarkers and specific criteria for severe disease for this species in the literature. A sine qua non requisite in the analysis of clinical severity related to P. vivax infection is the exclusion of mixed infection with P. falciparum through a more sensitive technique such as PCR and the exclusion of other co-morbidities which may be responsible for the clinical presentation per se. In the literature, in general, reports of 'complicated/severe' cases lack more precise and uniform definition criteria, in part due to the rare application of more robust endpoints such as death and admission to the intensive care unit (ICU), and therefore end up suffering bias through individual judgment of authors, editors and reviewers. As in most of the data published there were no systematic exclusion of co-morbidities and/or mono-infection confirmation using PCR, performing a meta-analysis of severe manifestations of P. vivax becomes virtually impossible. The other bias in the case of Brazil is that many relevant data are confined in abstracts from national scientific meetings and graduate students' dissertations and theses. The systematic review of these unpublished data therefore could contribute to the understanding of the clinical spectrum of vivax infection in this country and ultimately as a representative sample from Latin American vivax malaria. The sources for published data on clinical aspects of vivax infection in Brazil were MEDLINE (1948 to February 2011) and LILACS (1982 to February 2011). The following search strategy was devised for both databases: (Plasmodium vivax).mp. AND (Brazil).mp. All types of study designs with primary data were included (cross-sectionals, case-controls, cohorts, case series and case reports). The abstracts were analysed in details by two independent reviewers and publications were selected if they mentioned any type of clinical complication (no specific criterion was used) in at least one patient with the diagnosis of vivax infection. Disagreement between the two reviewers was solved through consensus. Articles were excluded if they were reviews and also if they did not contain primary data on clinical aspects. For included studies, there were extracted data on date of publication, location, number of patients, and characteristics of participants (age range, pregnancy status, presence of co-morbidities), if molecular diagnosis through PCR was used to assess vivax malaria mono-infection and fatality. Exclusion criteria for analysis were participants with mixed infections (P. falciparum/ P. vivax); studies in where patients with P. falciparum and P. vivax were both presented but the clinical data reported was not individualized for each species; and studies reporting the same patients from previous studies from the same authors. Through abstract analysis, 297 articles were retrieved and after application of the inclusion and exclusion criteria, 27 articles (from 1987 to 2011) were selected, which are presented in Table 1 . Unpublished studies were searched manually in the Annals of the Congress of the Brazilian Tropical Medicine Society (published in supplements of the indexed journal Revista da Sociedade Brasileira de Medicina Tropical [Journal of the Brazilian Society of Tropical Medicine]), from 1964 to 2011. This is the most traditional scientific event for tropical medicine clinicians in Brazil. Similar inclusion and exclusion criteria were used in this search. However if the same abstract data were published afterwards as a full paper, the published paper information was presented here. If the abstract referred to a dissertation or thesis, this more detailed information was presented instead. Forty-five abstracts were retrieved from 1995 to 2011. Of these, 17 fulfilled any of the exclusion criteria and therefore, 28 abstracts are presented in Table 2 . Masters' dissertations and doctoral theses abstracts since 1987 were searched in the online database http:// capesdw.capes.gov.br/capesdw/Teses.do maintained by the Coordination for the Improvement of Higher Education Personnel (CAPES), the institution which coordinates and supervises all the Brazilian Graduate Programmes in all areas of knowledge. The full original electronic documents were downloaded from the website when available or obtained through contact with the respective graduate students. Ten dissertations and three theses are presented in Table 3 . Classical malaria paroxysms are typically short and sharply delineated within a period of less than eight hours. Fever is one feature that is almost invariably present during a paroxysm. Any of other common symptoms of the febrile syndrome, such as chills, rigours and sweating, are also described. These symptoms of a paroxysm could be accompanied by others, including headache, nausea and vomiting, and moderate to severe muscle, joint and back pain [93] . Indeed high fever tends to be more evident in vivax disease even with lower parasitaemia, due to its recognized lower fever-threshold (around 100 infected RBCs/microlitre) [94] . Therefore, any description of these classical symptoms, together or isolated, should be regarded by any experienced clinical as non-severe malaria, regardless of their intensity, because they are not associated to increased rates of hospitalization or fatality. In the Brazilian literature reviewed, a wide spectrum of clinical complications aside from the classical symptoms of vivax malaria was found throughout the 68 indexed and non-indexed publications, despite the low number of deaths attributed to this species in this literature sample. The major complications are addressed as follows: World Health Organization (WHO) criterion for severe anaemia is haemoglobin below 5 g/dL in children and under 7 g/dL in adults. However the clinical manifestations due to anaemia per se are not known and to what extent it contributes to the respiratory distress associated with the hyperdynamic status of the febrile syndrome. There is scarce literature on malarial anaemia in population-based studies in Latin America, as reviewed elsewhere [95] . On top of that, major differences in Latin America are seen when the same methodology is applied. That is probably related to distinct genetic background and environmental factors, e.g. in the Amazon Basin (intense racial mixture) and in the Colombian Pacific Coast (nonmixed black population) [96] . It is not known if anaemia is as frequent among patients from Brazil as in Southeast Asian patients, where P. vivax is considered to be a disease of children because the acquisition of immunity against this species occurs much faster than for P. falciparum, in highly endemic areas [97] . In Brazil only 25% of vivax disease affects children 0-14 years of age, however severe anaemia was reported in hospitalized children and adults, needing red blood cell (RBC) transfusions [45] . A key description of anaemia in vivax malaria children in Latin America was published in Venezuela in 2006 [98] . The 'congenital malaria' in newborns from the present series of reports with severe anaemia confirms previous findings that vivax malaria has an important clinical impact in children under 3 months [99] . Non-severe anaemia, however, seems to be as frequent as 25.8% among the population of a recent occupation area in Rondônia, where hydroelectric power plants are being built [100] . The cut-off of haemoglobin under 12 g/dL as a criterion of anaemia however should be seen with scepticism because of age ranges and the lack of baseline levels of haemoglobin validated to specific populations, which makes meta-analyses susceptible to misclassification. Major confounding factors in the global analysis of anaemia are the local contributors to this haematological complication such as iron-deficiency anaemia, which was found to occur in 5.6% of a rural Amazonian population, mostly among school children and women [101] . Another important associated condition, which may interfere in the comparison between distinct populations, is the prevalence of intestinal helminthic infection. In a study performed with anaemic children, the presence of hookworms and malnutrition was cited [30] . However some controversy exists regarding this influence since in a cohort study, children with any intestinal helminth were protected from anaemia triggered by acute vivax infection [47] . In fact anti-helminthic treatment and iron supplementation reduced the haematological indexes in the population from an endemic area for malaria [102] . No Brazilian study has addressed the concomitant diagnosis of parvovirus B19 as a contributing factor to anaemia in malaria, considering that recent evidence supports that the use of chloroquine (CQ) may stimulate viral replication in the bone marrow, worsening anaemia [103] . Apparently pregnant women develop anaemia as a major complication in vivax infection [32, 78] , and the impact upon the concept needs further investigation. Chronic comorbidities affecting erythrocyte physiology, such as sickle cell anaemia (SCA), may be related to more severe haemolysis and severe anaemia as well [40] . Thrombocytopaenia as defined by platelet counts under 150,000/μL seems to be very frequent among patients with vivax malaria and apparently more frequent in vivax than in falciparum patients [104] , despite not being a consensus [105] . The increase in the report of thrombocytopaenia in several reference centres could also be a reflection of a better laboratorial infrastructure. Only in recent decades in developing countries automated full blood counts included platelet count as a routine. Many studies in Brazil confirm that platelet counts are directly correlated to peripheral parasitaemia [89, 90] , but the meaning of this finding is still unknown. However, only mild bleeding is usually associated with this haematological complication in studies where detailed and systematic clinical description of the patients was made, even for severe thrombocytopaenia, which means in general platelet count under 50,000/μL [81, 89, 90] . In fact, there is no report in the whole literature of a fatal case of patient presenting exclusively with severe thrombocytopaenia, even for P. falciparum. That is probably why thrombocytopaenia, regardless of being described as a complication by WHO, is not strictly-speaking considered a severity criterion by itself [106] . What happens most of the time is that thrombocytopaenia is usually taken as a surrogate marker for DIC in settings where no specific examinations to confirm this severe complication are available, such as prothrombin activation time, D-dimers and fibrin degradation products. However, there is a disproportionate difference in the proportions of thrombocytopaenia, which is considered relatively frequent in large studies for frequency estimation [90] and of DIC, which is a rare complication, very scarcely reported in the literature associated to P. vivax infection [107, 108] . Actually, there is some coagulation cascade activation, but usually with minor impact on coagulation tests and platelet counts [83] . It is important to consider however that in areas where dengue is also endemic, as is the case of Brazil, thrombocytopaenia studies should obligatorily rule out this viral infection, which also presents a substantial percentage of thrombocytopaenia as part of its non-severe presentation [109] . In fact there are cases of co-infection already reported in the Brazilian Amazon recently [110] , but the literature poorly describes the clinical aspects of this coincidental infection [111] . Respiratory distress is defined by oxygen saturation less than 94%, or deep breathing (acidotic breathing), or an age-stratified increased rapid respiratory rate (> 32/min in adults, > 40 in children 5-14 y, > 50 in children aged 2 mo to 5 y, and > 60 in babies less than 2 mo) [112] . However this syndromic approach does not translate any mechanism of disease and may be associated to the clinical presentation of febrile syndrome during the malarial paroxysm, severe anaemia, metabolic acidosis, lung oedema, pneumonia or acute respiratory distress syndrome (ARDS). In most of the cited Brazilian studies, there are no described criteria on how respiratory distress was defined, which makes comparisons with the general literature impossible. Sometimes imprecise clinical presentation is simply defined as pulmonary manifestations. In only one, ARDS is well characterized, comprising detailed radiological characterization and arterial gas analysis (FiO 2 /PaO 2 ) [37] . Lung oedema is usually based on clinical and radiological parameters and the effect of fluid overload is not clear for vivax infection, since only a few cases were reported so far, Brazilian cases included [39, 64, 65, 113, 114] . The impairment of respiratory symptoms after the beginning of treatment with CQ referred elsewhere [115] was not mentioned in any of the present reports, which could be due to inappropriate study design. Ruling out pneumonia is not easy because of the low frequency of positive blood cultures and due to the fact that in most of these patients with pulmonary complications empirical antibiotics are initiated as a rule. Data from the Papuan Indonesia indicate that many infants who die with P. vivax have radiological evidence of pneumonia [116] , but the specificity of radiological findings to differentiate vivaxinduced pulmonary abnormalities from pneumonia is questionable. In Mozambique, pyogenic bronchopneumonia was a common cause of respiratory distress in autopsied pregnant women with falciparum malaria, in both HIV positive and negative [117] . In the Amazon, HIV prevalence is estimated to be~1% (unpublished data), which makes opportunistic diseases less prone to impact on severe clinical complications of vivax malaria, as is the case for falciparum malaria in Africa. This is classically the most lethal clinical complication of severe falciparum malaria and the definition is also very imprecise with a wide spectrum of possible presentations, such as: impaired consciousness or unrousable coma (Glasgow coma score ≤10 or Blantyre coma scale ≤2); prostration, i.e. generalized weakness so that the patient is unable walk or sit up without assistance; failure to feed; or multiple convulsions (more than two episodes in 24 h). Despite being infrequent in our studies, the phenomenon was also reported but not only in children [36, 86] . These reports must be very cautious in terms of ruling out other malarial complications as the cause of the neurological manifestations, such as hypoglycaemia and metabolic acidosis, but also associated infections as bacterial or viral meningoencephalitis. In India, acute intermittent porphyria was an unexpected co-morbidity associated to the neurological manifestations of patients with vivax malaria [118] . In Papua, P. vivax-associated coma was rare, occurring 23 times less frequently than that seen with falciparum malaria, and was associated with a high proportion of non-malarial causes and mixed infections detected using PCR [119] . This complication is suspected in cases of oliguria and confirmed if serum creatinine is higher than 3.0 mg/dL. Bacterial sepsis, dehydration, shock and past history of chronic renal failure should be routinely searched in the differential diagnosis. It was also reported in the Brazilian literature [38, 48] , but in one study one case was found in a patient with arterial hypertension, what could be a triggering condition [45] . Despite not being frequent in Brazil, Plasmodium malariae is found in some scattered areas [120] , and as a potential cause of glomerulonephritis [121] , this parasite should be ruled out by molecular biology tools whenever acute renal failure is detected in a malarial patient with vivax infection, due to similarities of these two species at routine optical microscopy. New WHO guidelines already point to hyperbilirubinaemia (total bilirubin > 3.0 mg/dL) as being a weak marker of severity, unless it is followed by any other vital organ dysfunction [106] . This finding seems to be the most frequent among children and adults with vivax disease considered as 'severe' [122, 123] . Since haemolysis is not usually as severe as to cause significant clinical jaundice, most of these patients actually have some hepatocyte necrosis as evidenced by the mild to moderate liver enzymes (AST/ALT) increase with subsequent cholestasis [124] . It was shown that icteric syndrome was a common cause of hospitalization in pregnant women with vivax malaria in Manaus [87] . It was also detected in newborns [59, 74] , which makes vivax malaria an obligatory differential diagnosis of neonatal sepsis. Jaundice in the presence of vomiting and upper abdominal pain should raise suspicion on acalculous cholecystitis, a poorly described complication apparently with good prognosis [125] . Other diseases that may evolve to an icteric syndrome may be ruled out, especially because they are also more frequent in the tropics, such as leptospirosis [34] and typhoid fever [126, 127] . Hepatitis A virus (HAV) and vivax co-infection has already been reported as cause of jaundice and high elevation of transaminases [45] . Hepatitis B virus (HBV) is also highly prevalent in Brazil, especially in the Amazon [128] and there is some evidence that P. vivax/HBV co-infection may be related to more frequent jaundice [80] and higher transaminase levels. Algid malaria refers to the shock syndrome usually defined as circulatory collapse (systolic pressure under 70 mmHg in adults or under 50 mmHg in children) non-responsive to fluids. In the present vivax malaria reports, it was more reported most frequently among patients who died, suggesting that, as expected for this severe clinical complication, it could be regarded as a good marker of severity. However, the aetiology of this complication is still unclear even for P. falciparum. Apparently it is multifactorial and the complication should be regarded as a syndrome where cardiac dysfunction, dehydration, bleeding, adrenal insufficiency, and bacterial sepsis could all play a role [129] . A review of all malaria deaths in the USA found that 5% were due to P. vivax associated with cardiac disease [130] , which suggests cardiac dysfunction as a contributing factor to algid malaria. There is robust evidence that bacteraemia in Africa is associated with higher fatality in falciparum malaria in children [131] . Less frequently shock occurs isolated, but usually as part of multi-organ dysfunction syndrome (MODS), leading to a clinical picture suggestive of 'malaria-induced toxic shock' [132] . Metabolic acidosis (plasma bicarbonate < 15 mmol/L) and hyperlactataemia (lactate > 5 mmol/L), which are common in severe falciparum malaria and are good predictors of fatal outcome, have never been described in vivax severe disease. In a series of children with vivax infection admitted to the ICU, metabolic acidosis is mentioned [92] , however concomitant sepsis is described in this series and specificity for malaria cannot be assumed. If one admits lactic acidosis as a consequence of hypoxia triggered by microvasculature obstruction in falciparum disease, the scarcity of data on the frequency of this phenomenon in vivax disease may simply reflect the less severe obstruction due to less cytoadhesion, as already suggested elsewhere [133, 134] . In the case of hypoglycaemia (blood glucose < 40 mg/dL), the complication has been rarely described elsewhere [123, 135] , and in only two studies in Brazil this finding was reported among children and pregnant women [32, 92] . The impact of vivax infection upon pregnancy and the concept is less clear in Brazil and Latin America as a whole, despite robust evidence that vivax malaria causes low birth weight and maternal anaemia exists in Thailand [136] and Indonesia [137] . The burden of the infection due to this species in Brazilian pregnant women from a highly endemic area in the Amazon seems to be high [138] . Malaria anaemia in pregnant women with vivax is already known [137] and data from Brazil confirm that this is the most common complication among these women [78] . Additionally the few reports in the present series also point to low birth weight, vaginal bleeding, amniorrhexis, abortion, premature delivery, hypoglycaemia, hepatitis and jaundice as complications [31, 32, 49, 70, 87] . Hyperemesis gravidarum may superimpose to the febrile syndrome and to the gastrointestinal side effects associated to CQ in pregnant women, contributing to uncontrolled vomiting and consequent metabolic disorders. Apparently in the case of pregnancy, co-morbidities do not seem to be frequent among patients with clinical complications. Despite the need of more pathogenesis studies with the infected placenta, ultrasound studies in order to search for prognostic markers are urgently needed. Some atypical complications are not frequently described for malaria and likewise are not classically referred as severe malaria. Rhabdomyolysis has been reported for vivax in 1993 in a patient with myoadenylate deaminase deficiency [139] ; only one case was reported in Brazil in a patient without co-morbidities [46] . Rarely, patients with vivax malaria could evolve with immune thrombocytopenic purpura (ITP) as a complication of the acute infection [33] . To confirm this diagnosis, the patient has to be followed up with persistent thrombocytopaenia for many weeks after the efficacious anti-malarial treatment and diseases, in which ITP is more frequently seen, such as HIV, should be discarded. The mechanisms involved are poorly understood. Splenomegaly is considered a typical finding in the physical examination of a patient with vivax disease, but the occurrence of spleen haematomas evolving with rupture and fatal outcome is relatively rare [41, 52, 91] despite being more frequent among this species as compared to falciparum [140] . In any case, patients with vivax malaria referring abdominal pain should be investigated for this complication as some patients may evolve with a bad prognosis if not properly managed by a surgeon. Ocular manifestations in vivax disease apparently have no relation to cerebral malaria or bad prognosis, as is the case for falciparum [141] . Few reports have been published on vivax patients with non-severe disease and retinal haemorrhage [142] and in Brazil this fundoscopical finding was associated with hypovitaminosis A [35] . Another atypical complication, which may be more frequent than expected for vivax infection, and with outstanding impact upon the development of some emerging economies in the globe, is poor school performance which should be a surrogate marker for the intellectual impairment related to malaria [143] . Acute malnutrition has been shown to be a complication of vivax malaria in highly endemic areas [144] . In Brazil a few evidences show that malnourishment and vivax co-exist but the impact of this association is still unknown [145] . In only two studies was malnutrition referred to as a possible cause of the reported clinical complication [92, 146] . Vasculitis [66] , leukemoid reaction [76] and pleural effusion [62] as a marker of severity seems to be speculative and details of these reports do not support any in-depth analysis. High parasite density as a marker of severity for P. vivax, as it is for P. falciparum, still needs additional studies, considering this parasite infects preferably reticulocytes. The same occurs with the presence of schizonts in peripheral blood, which is usually associated with high sequestered biomass and severity for falciparum [147] , but is still an unexplored aspect for vivax. Strong linear trends were identified regarding increasing plasma levels of C reactive protein (CRP) and the gradation of disease severity [48] . Super-oxide dismutase-1 (SOD-1) seems to be a powerful predictor of disease severity in individuals with different clinical presentations of vivax malaria [148] . As soon as precise markers of severity are available, it would be possible to design studies powered to analyse the influence of the host genetics in the development of severe vivax disease. Some association between pulmonary manifestations and TNF and IL-12 polymorphisms has been attempted [79] . It has been proposed for the first time in Manaus that G6PD deficiency could protect against vivax malaria, in a cross-sectional study, based on past history of the enrolled population [149] . This protection was later confirmed in Pakistan [150] . Male hemizygotes for this deficiency also showed to be protected against severe falciparum malaria [151] . No data exist on the protection against severe vivax disease. Likewise, people with the FYA/FYB genotype presented higher susceptibility to clinical vivax malaria [152] . Since the discovery in Brazil that Duffy-negative individuals could be infected by P. vivax [153] , some speculation on the other possible invasion receptors has emerged. However, cohort studies are needed to investigate the real impact of the distinct Duffy genotypes on clinical malaria incidence, submicroscopic asymptomatic infection, malaria-triggered anaemia and lower parasitaemia, as already suggested for FYB/FYX and FYA/FYX genotypes in the Brazilian Amazon [154] . The major advance in the study of the pathogenesis of severe vivax disease was the demonstration of P. vivaxinfected RBCs cythoadhesion on human lung endothelial cells (HLEC) and placental tissue ex vivo [134] . This cythoadhesion was obviously lower than P. falciparuminfected RBCs adhesion, but with similar stability. However the next challenge is to try to link this finding to the in vivo phenomena [155] . The increased adhesion with the addition of LPS in the P. vivax ex vivo model suggests that endothelial activation may be an enhancing event. The role of augmented platelet-derived microparticles [156] and CD4 + CD25 + FoxP3 + regulatory T cells (Tregs) cells found in vivax disease should also be investigated in severe disease. Plasma levels of TNF, IFN-γ and also IFN-γ/IL-10 ratios were increased and exhibited a linear trend with gradual augmentation of disease severity [48] . Patients with severe disease also presented higher haemolysis and higher plasma concentrations of Cu/Zn SOD-1 and lower concentrations of PGE-2 and TGF-β than those with mild disease [157] . Oxidative stress was also proposed as a mechanism for thrombocytopaenia found in vivax disease [158, 159] , as well as its association with TNF [85] . Circulating immune complexes were not associated to vivax thrombocytopaenia [89] , but polymorphisms of the highly immunogenic AMA-1 were associated to platelet count in these patients [160] , suggesting that immunological mechanisms are involved in platelet destruction. In the case of anaemia, there is no correlation between the presence of anti-erythrocyte and anti-cardiolipin antibodies and the presence or intensity of this haematological finding [161] . Auto-immunity induced by secondary cryoagglutinins should be explored [54] . Erythropoiesis seems to be affected [162] , and the finding of parasites inside the bone marrow [43] stimulate the search for mechanisms of diserythropoiesis in this milieu, despite technical limitations to analyse this tissue in humans. The role of the spleen in severe disease is still unknown, as well as the role of the variant subtelomeric multigene vir family, which may influence the sequestration of infected RBCs in this organ [163] . Parasite genetics, such as MSP-1 and CSP polymorphisms, has not been shown to be associated with clinical severity [164] . In summary, the immune response in patients with severe vivax disease has not been fully addressed in the general literature, and further approaches are needed in order to unveil immune mechanisms related to these complications. In terms of therapy, CQ and primaquine (PQ) are still the drugs of choice for the treatment of vivax malaria in many endemic areas, Brazil included. It is important however to keep in mind that side effects of these drugs could be erroneously taken as clinical severity associated to the parasite infection. In the case of CQ, it is considered a safe drug, despite the occurrence of pruritus, which most of the time is considered to be a minor effect and rarely requires the drug withdrawal [165] . Psychosis on the other hand is a more severe complication [75, 166] , as well as cardiac arrhythmia [29] . Atypical complications of its use such as severe gastric bleeding were associated with haemophilia A [50] . In the case of PQ, tranquillity is not the same as with CQ, because PQ is able to induce metahaemoglobinaemia [42, 51] and severe haemolysis [44] in patients with G6PD deficiency. The burden of the deficiency in Brazil is poorly measured but the few data available in endemic areas for malaria has shown it to be between 3.0% [149] and 5.8% [167] among men, since the deficiency is linked to the X-chromosome. In the case of Brazil, the prescription of PQ in the abbreviated regimen (0.5 mg/kg/day for 7 days) without any routine G6PD screening may contribute to increase the frequency and severity of the side effects triggered by this drug, as confirmed by the reports of patients with blackwater fever after PQ use, including one fatal case [63, 71, 91, 92, 168] . To complicate matters, for the radical cure, the new drug under late stage clinical investigation, tafenoquine, shows no evidence that it is safer than PQ in G6PD deficient [169] . The simultaneous occurrence of severe vivax disease and CQ-resistance in some countries has raised the question of a possible association between severity and resistance, especially for anaemia [170] . CQ resistance actually has been reported in Brazil almost at the same time as clinical severity [171, 172] , but some studies argue against that, showing that severe patients responded to CQ [45] . Added to that, reliable genetic markers of resistance are lacking [173] . Increased levels of pvmdr-1 and pvcrt-o RNA in a single severe patient with vivax malaria however paved the way to the study of gene expression in association to resistance [174] . As suggested by the present data, 11 cases were reported in Brazilian travellers who live in the non-endemic area and occasionally go to the Amazon. Regarding the possibility of severe disease triggered by P. vivax, Travel Outpatient Clinics should emphasize to their clients the possible complications of this disease, still considered 'benign' in most of the educational folders and travellers' guides, especially because no good chemoprophylaxis against relapses related to this species is available to date. On top of that, retarded diagnosis and treatment outside the Amazon area contributes to the higher fatality rate of P. falciparum patients [18] . A similar situation could be observed for P. vivax, being this disease misdiagnosed as other febrile diseases. Despite the increasing evidence of CQ-resistance worldwide, the Brazilian Ministry of Health still recommends CQ as the first line therapy for vivax treatment, considering that only one single study has properly shown~10% of resistance in the area of Manaus [172] . The few available efficacy studies on ACT for the treatment of vivax were reviewed recently [175] , and give good evidence for their use in vivax malaria, however, more studies are needed. Only recently the Brazilian Ministry of Health followed the WHO recommendations to manage vivax severe patients with parenteral artemisinin derivatives as if they had severe falciparum infection, considering that a submicroscopic mixed infection could be misdiagnosed in the routine TBS [106] . This recommendation was already stated by the famous Brazilian parasitologist Samuel Pessôa in his Medical Parasitology textbook, from 1967 [176] . Supportive therapy is even more neglected and there is virtually no study focusing in the clinical management of patients with severe vivax disease. There are actually many priorities in clinical research related to vivax disease. The major ones were discussed previously. Considering that asymptomatic infections due to P. vivax are even more common in endemic areas for both species [177] , the likelihood of an asymptomatic patient becoming ill due to another microorganism is not improbable, which requests a good epidemiological characterization of the endemic area where the severe cases are being reported and systematic exclusion of mixed infections through PCR, due to the possibility of submicroscopic infection with P. falciparum. Another major priority in vivax research is the investigation of concurrent infections through systematic laboratory exclusion of the most prevalent infectious diseases in severe patients. In Figure 1 , the major research questions are addressed. In the present systematic review, the major limitation was the fact that most of the information was retrieved from non-peer reviewed sources. However, it seems clear that vivax patients in Brazil are calling the attention of their physicians only recently. Like other infectious diseases, defining severity criteria is a major challenge. As an example, dengue fever specialists have defined 'warning signs' for dengue haemorrhagic fever, the most lethal complication of the infection due to dengue virus, which are early signs that should raise the suspicion of severe dengue but are not applied themselves to the final classification [178] as proper intervention can avoid the patient evolving to more severe stages. Sometimes in the literature potential 'warning signs' for severe vivax malaria are mistaken for severity criteria, which are those ultimately related to increased fatality. WHO severity criteria formerly developed for falciparum disease seem to apply reasonably to vivax disease as well, but there are clearly 'warning signs' that should motivate clinicians from the tropics to observe patients more closely, such as isolated thrombocytopaenia, isolated jaundice or the presence of chronic or acute co-morbidities. For example, during influenza outbreaks, the virus does not necessarily kill per se, but compromises the most vulnerable population and facilitates fatal secondary bacterial infections. The most common complications observed in the field are not necessarily the most frequently reported in the literature, sometimes biased by the uniqueness or exoticness of the cases reported. It is only after 1987 that these cases started to be reported in Brazil in indexed and nonindexed publications, which may simply parallel the increase in the absolute numbers of vivax cases in Brazil, culminating in the more frequent observation of rare clinical events triggered by this parasite. Publication bias may also impact the chronology of these complicated case reports, especially when research group leaderships based in the endemic areas start to look for clinical aspects more closely. It is noteworthy however that studies on pathogenesis must be careful when dealing with severe vivax disease as a single entity. The best approach is to study groups of patients with specific complications (e.g., severe anaemia or ARDS) in order to minimize the risk of heterogeneous groups with probable multifactor causality, including the diversity of host genetics. The amount of complications related to anti-malarial drug use is not negligible, especially primaquine. Multicentric studies using standard protocols, with the proper care of confirming mono-infection by more specific tools (e.g. PCR) and ruling out co-morbidities, are urgently needed to characterize the real spectrum of vivax disease worldwide. Tissues from deceased patients are also waited, in order to support more robust analyses of the mechanisms of death. Without that information, vaccine clinical trials against P. vivax will not be able to include among their endpoints the protection against the severe disease (essentially severe anaemia), which parallels the frequency of severe falciparum anaemia in some endemic areas. The recent discussion on malaria eradication will only succeed if the two parasites which most affect humans begin to be treated as distinct and not causing a single disease. Clinical characterization is the first step to estimate its burden and ultimately to plan any control strategy in the near future.

Protective Role of the ACE2/Ang-(1–9) Axis in Cardiovascular Remodeling

Despite reduction in cardiovascular (CV) events and end-organ damage with the current pharmacologic strategies, CV disease remains the primary cause of death in the world. Pharmacological therapies based on the renin angiotensin system (RAS) blockade are used extensively for the treatment of hypertension, heart failure, and CV remodeling but in spite of their success the prevalence of end-organ damage and residual risk remain still high. Novel approaches must be discovered for a more effective treatment of residual CV remodeling and risk. The ACE2/Ang-(1–9) axis is a new and important target to counterbalance the vasoconstrictive/proliferative RAS axis. Ang-(1–9) is hydrolyzed slower than Ang-(1–7) and is able to bind the Ang II type 2 receptor. We review here the current experimental evidence suggesting that activation of the ACE2/Ang-(1–9) axis protects the heart and vessels (and possibly the kidney) from adverse cardiovascular remodeling in hypertension as well as in heart failure.

All epidemiological studies show that the risk of adverse cardiovascular (CV) outcomes, such as stroke, myocardial infarction (MI), heart failure (HF), and kidney disease [1] , increase progressively with increasing blood pressure (BP). On the other hand, clinical trials demonstrate that lowering BP reduces such risks [1] . All antihypertensive medications lower BP, but specific drug classes display effects beyond BP reduction (pleiotropic effects) that might contribute to cardiovascular risk reduction. Remodeling of the cardiovascular structure occurs in response, not only to changes in BP and flow, but also to modifications in the neurohormonal environment, in which the rennin-angiotensin-aldosterone system (RAAS) exerts a most predominant influence [2] . The RAAS is a major regulator of BP [3, 4] . In addition, the RAAS has a role in the vascular response to injury and inflammation [4] . Chronic RAAS activation, through both angiotensin (Ang) II and aldosterone, leads to hypertension and perpetuates a cascade of proinflammatory, prothrombotic, and atherogenic effects associated with endorgan damage [3, 4] . Based on these facts, several drugs have been developed that work by (a) reduction of Ang II levels, (b) inhibition of the Ang II type 1 receptor (AT1R), (c) blockade of the aldosterone receptor, and (d) renin receptor blockade [5, 6] . During the last 25 years several clinical trials have shown the benefits with these drugs that inhibit the RAAS with regard to BP reduction, regression of cardiac hypertrophy, prevention of kidney damage and reduction of cardiovascular morbidity reduction in hypertensive patients. Besides, with most of these RAAS blockers, quality of life as well as survival has been significantly improved in patients with heart failure. Consequently, the RAAS is currently a main therapeutic target in hypertension treatment [3, 4] . Aggressive BP control improves outcomes in patients with CV disease, stroke, and nephropathy and might have beneficial effects beyond BP lowering [7] . Despite the reduction of CV events and end-organ damage with the current pharmacologic strategies, CV disease remains the primary cause of death in the world, and more than 94,000 Americans annually experience progression to end-stage renal disease (ESRD). As population ages, the proportion affected by end-organ damage is expected to grow [8] . Thus, it is most relevant to find new molecules 2 International Journal of Hypertension in order to prevent and reduce hypertension as well as pathologic CV and kidney remodeling and dysfunction. In this regard, activation of the new ACE2/Ang-(1-9) pathway seems to counterbalance the damage due to the RAAS system activation. We review here the current experimental evidence suggesting that activation of the ACE2/Ang-(1-9) pathway protects the heart and vessels (and possibly the kidney) from adverse cardiovascular remodeling in hypertension as well as in heart failure. The discovery of angiotensin-converting enzyme homologue, ACE2, added further complexity to the main axis of the RAAS, in which Ang II and its forming enzyme ACE play major roles [9, 10] . A growing body of evidence points to a possible promising role for this new member of the RAAS by opposing to the effects of the main axis [11, 12] . ACE2 has dramatically changed the direction of cardiovascular and renal research in view of the pivotal role of this enzyme in the regulation of the RAAS [12, 13] . ACE2 is the newest member of the RAAS and shares approximately 40% similarity with the somatic form of ACE [9, 10] . ACE2 is a membrane-bound carboxypeptidase and its cellular and tissue distribution is different from that of ACE. While ACE is expressed in the endothelium throughout the vasculature, ACE2 is distributed in tissues with the most abundant expression in heart, kidney, lung, small intestine, and testis [14] . ACE2 can be released into the circulation and urine by shedding [15] . Tumor necrosis factoralpha-converting enzyme (TACE/ADAM17) is the sheddase responsible for the ectodomain cleavage and shedding of ACE2 [16] . However, normal ACE2 enzymatic activity in plasma is very low, probably due to the presence of an endogenous inhibitor [17] [18] [19] . ACE2 is different from ACE in both substrate specificity and functions [9, 20, 21] . ACE2 can form (a) Ang-(1-7) through hydrolysis of Ang II and (b) Ang-(1-9) through hydrolysis of Ang I. This last reaction is negligibly slow and is several hundred times slower than Ang II hydrolysis by ACE2 to form Ang(1-7)-a vasodepressor peptide counterbalancing the vasopressor effect of Ang II [20, 21] . Ang-(1-7) can be subsequently converted to Ang-(1-5) by ACE [9, 20] or by neutral endopeptidases [9] , while Ang-(1-9) may be converted to Ang-(1-7) by ACE [9] . There is little evidence proving the existence of alternative hydrolysis of Ang-(1-9) to Ang II in some tissues. Drummer et al. [22] proved that homogenates of rat kidney, and in a lesser extent of lung, convert Ang-(1-9) to Ang II due to an ACEindependent aminopeptidase and N-like carboxypeptidase. Singh et al. [23] confirmed that the pathway Ang I-Ang-(1-9)-Ang II really exists in glomeruli of streptozotocininduced diabetes mellitus rats. Moreover, in human heart tissue the main products of Ang I degradation are both Ang-(1-9) and Ang II generated by heart chymase, ACE and a poorly identified carboxypeptidase A [24] . Although the data proving the existence of alternative pathways of Ang II production, in clinical practice we can still block only ACE or AT1R. ACE2 does not act on bradykinin metabolism and its activity is not inhibited by classic ACE inhibitors (ACEIs) [9] . Thus it has been proposed that ACE2 activity may counterbalance the effects of ACE by preventing the accumulation of Ang II in tissues where both ACE2 and ACE are expressed [25, 26] . ACE2 has several biological substrates and it is considered a multifunctional enzyme. Acting as a monocarboxypeptidase, it cleaves several other non-RAAS peptides which have roles in maintaining cardiovascular homeostasis such as (des-Arg9)-bradykinin, a member of the kininogen-kinin system [13] . (des-Arg9)-Bradykinin is formed from bradkinin by the action of carboxypeptidases and is an agonist of the B1 receptor, which is induced after tissue injury [27] . Bradykinin, a vasodilator which acts through the B2 receptor, is produced from its precursor kininogen by kallikrein and is degraded by ACE [13] . While, degradation of bradykinin by ACE is known to be an important aspect of BP regulation, the significance of the degradation of (des-Arg9)-bradykinin by ACE2 remains to be established. In addition to (des-Arg9)-bradykinin, ACE2 is also able to degrade apelin-13, a peptide proposed to cause vasoconstriction and known to regulate fluid homeostasis, and other non-RAAS peptides such as kinetensin, dynorphin A and neurotensin [20] . For a long time, Ang-(1-7) was thought to be devoid of biological activity, in spite of early reports on biological effects [28] . The importance of Ang-(1-7) was emphasized by the discovery of ACE2. Ang-(1-7) has been shown to release vasopressin as effectively as Ang II from neurohypophyseal explants [28] and to have actions opposing those of Ang II, namely vasodilation, antitrophic effects and implications of vasodilation caused by bradykinin [29, 30] . Several experiments suggest an important interaction between Ang-(1-7) and prostaglandin-bradykinin-nitric oxide (NO) systems. Ang-(1-7) binds to the Mas receptor (G protein-coupled receptor) which mediates vasodilating and antiproliferative actions of this peptide [31] . The Mas receptor can hetero-oligomerize with the AT1 receptor and acts as a physiological antagonist of Ang II [32] . Studies revealed that Ang-(1-7) activated endothelial nitric oxide synthase and NO production via Akt-dependent pathways [33] . Furthermore, Tallant et al. [34] showed that the presence of an antisense probe directed against Mas abolished the Ang-(1-7)-induced inhibition of protein synthesis in cardiomyocytes. This study also revealed that Ang-(1-7) decreased serum-stimulated ERK1/ERK2 mitogen-activated protein kinase activity, a response that was blocked by D-Ala 7-Ang-(1-7), an antagonist of Mas receptor. Ang II binds with high affinity to two different receptor subtypes-AT1R and AT2R-which are members of the seven-transmembrane-domain G-protein-coupled receptors (GPCR) superfamily, through Gq and Gi, respectively [35] . Whereas the AT1R mediates most of the recognized actions of Ang II, it appears that the AT2R opposes, in part, to the effects mediated by the AT1R. As the AT2R is expressed in adult tissues in smaller amounts than the AT1R, the actions and cell signaling of AT2R have been less well characterized International Journal of Hypertension 3 than those of AT1R [36] [37] [38] . Current knowledge suggests that AT2R stimulation mediates vasodilation, antigrowth, proapoptotic and antiinflammatory effects [39, 40] . Hence, the AT2R can modulate cardiovascular remodeling as well as progression of atherosclerosis. AT2R stimulation activates the NO-cGMP-dependent pathway [41] . This occurs either directly or indirectly through bradykinin or by increased endothelial NOS activity or expression. AT2R activation is associated with phosphorylation of JNK, PTPs, IκBα (inhibitor of NF-κB), and the transcription factor ATF2, and dephosphorylation of p38MAPK, ERK1/2, and STAT3, which are linked to antiproliferative and antiinflammatory effects and apoptosis [38, [42] [43] [44] . AT2R may induce relaxation by opening large-conductance Ca 2+ -activated K + channels (BKCa) [45] and by negative regulation of the vascular Rho A/Rho kinase pathway. The AT2R also enhances the activity of tyrosine phosphatases and vanadate-sensitive phosphatases MKP1 (DUSP1), SHP1 (PTPN6) and PP2A [46, 47] . There is little information in the literature with respect to Ang-(1-9) probably because this peptide was initially thought to be active only after conversion to Ang-(1-7). Ang-(1-9) can be generated by several carboxypeptidase-type enzymes including ACE2 or cathepsin A [48, 49] . Ang-(1-9) is present in healthy volunteers, in patients or in animals treated with ACE inhibitors (ACEIs) or AT1 receptor blockers (ARBs) [50] [51] [52] , and its circulating levels are increased by pathological conditions (i.e., early after MI) [51] . However, very little is currently known about Ang-(1-9) biological effects [50, 53] . Initial studies showed that incubation of Chinese hamster ovary cells (CHO) with Ang-(1-9) potentiated the release of arachidonic acid by [Hyp 3 Tyr(Me) 8 ]BK, elevated [Ca 2 ]i and also resensitized the B2 receptor desensitized by BK [48] . At the same time, Jackman et al. [54] showed in CHO cells and in human pulmonary endothelial cells that Ang-(1-9) was significantly more active than Ang-(1-7) enhancing the effect of an ACE-resistant bradykinin analogue on the B2 receptor and that Ang-(1-9) also augmented arachidonic acid and NO release by kinin [54] . Some studies have suggested that Ang-(1-9) may be an endogenous inhibitor of ACE. Donoghue et al. [9] proposed that Ang-(1-9) is a competitive inhibitor of ACE because it is by itselfan ACE substrate. Under conditions of ACE inhibition, such as after long-term administration of an ACEI in rats, Ang-(1-9) levels increased in plasma and kidney [50, 53] . This increase in Ang-(1-9) steady-state levels could be due to decreased catabolism of Ang-(1-9) by ACE. Conversely, the increased levels of Ang-(1-9) could be due to increased production by ACE2 as a result of increased availability of Ang I substrate. These results indicate that an alternate pathway of Ang I metabolism by ACE2 exists and that this pathway may be amplified in the presence of ACE inhibitors. To determine whether Ang-(1-9) is active per se or it becomes active only after conversion to Ang-(1-7), Chen et al. [55] examined the metabolism of Ang I, Ang-(1-9) and Ang-(1-7) in stably transfected CHO cells that express human ACE and human bradykinin B2 receptors coupled to green fluorescent protein (B2GFP). They found that Ang-(1-9) was hydrolyzed 18 times slower than Ang I and 30% slower than Ang-(1-7). Ang-(1-9) inhibited ACE and it resensitized the desensitized B2GFP receptors, independently of ACE inhibition [55] . This is reflected by release of arachidonic acid through a mechanism involving cross-talk between ACE and B2 receptors. They concluded that Ang-(1-9) enhanced bradykinin activity, probably by acting as an endogenous allosteric modifier of the ACE and B2 receptor complex. Therefore, when ACE inhibitors block conversion of Ang I, other enzymes like ACE2 can still release Ang I metabolites like Ang-(1-9) and enhance the efficacy of ACEIs. Recently, Flores-Muñoz et al. [56] using radioligand binding assays observed that Ang-(1-9) is able to bind the Ang II type 2 receptor (AT2R) (pKi = 6.28 ± 0.1). They demonstrated that Ang-(1-9) and not Ang II, affected hypertrophy through the AT2R, as PD123319 (an AT2 receptor blocker) did not alter Ang II-mediated growth but did block the effects of Ang-(1-9). Despite having ∼100fold lower affinity than Ang II for the AT2R [57] , the selective AT2R activity of Ang-(1-9) is not inconsistent with current pharmacological models of G protein-coupled receptor signalling and activation. Indeed, the concept of functional selectivity, where individual receptor ligands have the capacity to selectively stabilize conformations which lead to distinct signalling outcomes [57] [58] [59] , is supported by a previous study in which the critical amino acids and the mode of binding of ligands at the AT1R and AT2R were investigated [60] . While agonist activation of the AT1R was particularly sensitive to peptide modifications that disrupted contact points between Ang II and its receptor, substitutions within Ang II were far better tolerated by the AT2R [60] . The AT2R exists in a relaxed conformation and Ang II therefore binds to multiple indistinct contact points [60] . Since Ang-(1-9) contains the entire Ang II sequence plus a C-terminal histidine, these observations indicate that this difference may stabilize the AT2R in a conformation able to counteract hypertrophic signalling in cardiomyocytes. Flores-Muñoz et al. [56] did not observe functional competition between Ang II and Ang-(1-9) at the AT2R and they concluded that that Ang-(1-9) is able to antagonize Ang II signalling in cardiomyocytes selectively via the AT2R, highlighting that Ang-(1-9), along with Ang-(1-7), makes up part of the counter-regulatory arm of the RAS. What remains to be determined is the downstream signalling effects from Ang- (1) (2) (3) (4) (5) (6) (7) (8) (9) . Preliminary studies indicate that the classical pathways via PKC translocation and ERK1/2 activation [61] [62] [63] are not different between Ang II-, Ang-(1-7)-and Ang-(1-9) stimulated cells. Since the downstream signalling from the AT2R is unclear at present, future studies will be required to establish these mechanisms. Crackower et al. [64] were the first to test ACE2 as the gene underlying the blood pressure locus on the X chromosome. They showed reduced expression of renal ACE2 in the salt-sensitive Sabra hypertensive rat compared with the normotensive rat. Both spontaneously hypertensive rats (SHR) and spontaneously hypertensive stroke-prone rats (SHRSP) rats showed reduced renal ACE2 protein levels compared with the normotensive Sabra and Wistar Kyoto (WKY) strains. Two other groups confirmed some of these findings showing lower renal ACE2 mRNA, protein, and activity in the SHR compared to WKY rats [65, 66] . However, other investigators were unable to detect any difference in renal ACE2 mRNA, protein, and activity between adult hypertensive rats and their normotensive controls [67] . Rentzsch et al. [68] , assessed in SHRSP (that display reduced ACE2 mRNA and protein expression compared with control animals in the kidney) the role of ACE2 in the pathogenesis of hypertension. They generated transgenic rats on a SHRSP genetic background expressing the human ACE2 in vascular smooth muscle cells by the use of the SM22 promoter, called SHRSP-ACE2. In these transgenic rats, vascular smooth muscle cells (VSMC) expression of human ACE2 was confirmed by RNase protection, real-time RT-PCR, and ACE2 activity assays. Transgene ACE2 expression leads to significantly increased circulating levels of Ang- (1-7) , a prominent product of ACE2. Mean arterial blood pressure was reduced in SHRSP-ACE2 compared to SHRSP rats, and the vasoconstrictive response to intraarterial administration of Ang II was attenuated. The latter effect was abolished by previous administration of an ACE2 inhibitor. To evaluate the endothelial function in vivo, endotheliumdependent and endothelium-independent agents such as acetylcholine and sodium nitroprusside, respectively, were applied to the descending thoracic aorta and blood pressure was monitored. Endothelial function turned out to be significantly improved in SHRSP-ACE2 rats compared to SHRSP. These data indicate that vascular ACE2 overexpression in SHRSP reduces hypertension probably by local Ang II degradation and by improving endothelial function [68] . A target gene therapy strategy holds significant potential to translate the available fundamental research of ACE2 into therapeutics. In fact, initial animal experiments have been extremely encouraging. For example, in SHR, viralmediated ACE2 overexpression in the heart decreased high BP [69] . This strategy also preserved cardiac function, as well as left ventricular wall motion and contractility, and attenuated left ventricular wall thinning induced by myocardial infarction [70] . ACE2 overexpression in the rostral ventrolateral medulla causes significant decreases in BP and heart rate (HR) [71] . Compared with ACEIs and ARBs, the targeting of ACE2 has the following potential therapeutic advantages, first, it degradates both Ang I to generate Ang-(1-9) and Ang II to generate Ang- (1-7) . Thus, targeting ACE2 would not only produce the antihypertrophic peptide Ang-(1-9) [52] and the vasoprotective/antiproliferative peptide Ang-(1-7) [72] [73] [74] , but would also influence the vasoconstrictive/proliferative effects of the ACE/Ang II/AT1R axis [75] . Second, it is a multifunctional enzyme with many biologically active substrates [9, 20] . Third, unlike ARB/ACEI therapy, ACE2 is an endogenous regulator of the RAS [75] . Fourth, it is a part of the vasodilatory/antiproliferative axis of the RAS [20] and fifth, although treatment with ACEIs or ARBs indirectly increases ACE2 expression, direct activation of this enzyme could result in a better outcome in cardiovascular diseases [68, 75] . Thus, the activation of the ACE2 axis may be a novel therapeutic strategy in hypertension. So far, all attention has been focused on Ang- (1-7) , that opposes the pressor, proliferative, profibrotic, and prothrombotic actions mediated by Ang II [76] . Experimental and clinical studies have demonstrated a role for the Ang-(1-7)/ACE2/Mas axis in the evolution of hypertension, the regulation of cardiovascular and renal function, and the progression of cardiovascular and renal disease including diabetic nephropathy [77] . Additional evidence suggests that a reduction in the expression and activity of this vasodepressor component may be a critical factor in mediating the progression of cardiovascular and renal disease. These findings support a role for the Ang-(1-7)/ACE2/Mas axis and, in particular, on its putative role as an ACE-Ang II-AT1 receptor counter-regulatory axis within the RAS [76, 77] . Recently, the alternative angiotensin peptide, Ang-(1-9) has shown relevant biological functions. Ocaranza et al. [51] have observed increased ACE2 activity and Ang-(1-9) plasma levels in MI and sham rats treated with enalapril for 8 weeks while circulating Ang-(1-7) levels did not change in any phase after MI [51] (Figure 1 ). These findings support the hypothesis that, in this second arm of the RAS, ACE2 through Ang-(1-9) instead of Ang-(1-7), could act as a counterregulator of the first arm, where ACE catalyzes the formation of Ang II. Besides, in experimental hypertension (DOCA salt model) and in normotensive sham animals, RhoA/Rho-kinase inhibition (a signaling pathway that participates in pathological cardiovascular and renal remodeling and also in blood pressure regulation) by fasudil reduced BP and increased vascular and plasma ACE2 enzymatic activity. At the same time, fasudil reduced Ang II and increased Ang-(1-9) plasma levels ( Figure 2 ) [78] . No modifications were observed here in Ang-(1-7) levels despite increased ACE2 levels with RhoA/Rho-kinase inhibition [78] . Thus, RhoA/Rho-kinase inhibition, by increasing eNOS and/or by reducing both ACE and Ang II, does not activate the Ang-(1-7) pathway. This novel effect of RhoA/Rho-kinase inhibition on both ACE2 expression and Ang-(1-9) levels might additionally contribute to the antihypertensive effects of RhoA/Rhokinase inhibitors. Besides, these results strongly suggest that in this experimental model, hypertension is more dependent on ACE2 and Ang-(1-9) levels than on ACE and Ang II levels. Therefore, this second RAAS axis through ACE2 and Ang-(1-9) could be an important target for the treatment of hypertension. International Journal of Hypertension 5 Figure 1 : Plasma levels of Ang-(1-7), Ang-(1-9) and Ang II in rats with myocardial infarction treated with the ACE inhibitor enalapril (8 weeks). Increased plasma levels of Ang-(1-9) were observed in rats with myocardial infarction treated with the ACE inhibitor enalapril. Myocardial infarction was induced by coronary artery ligation. Data are presented as mean ± SEM (n = 12/group). AMI: acute myocardial infarction, E: enalapril. * P < 0.05 compared to both Sham and untreated myocardial infarction groups; * * P < 0.05 compared to both Sham and enalapril-treated myocardial infarction groups. (adapted with permission from [51] ). The vascular wall is continuously exposed to hemodynamic forces such as the luminal pressure and shear stress. Changes in these forces, either physiological or pathological, lead to functional and/or structural alterations of the vascular wall [79] . Acute changes in hemodynamic forces can modify vessel diameter. Chronic changes in hemodynamic forces result in structural alterations of the vessel wall, indicated by changes in wall diameter and thickness. In addition, changes in vascular structure are not solely determined by hemodynamic forces [80] , but also by inflammatory responses and changes in extracellular matrix components [81] . Structural changes of the medial layer of the vascular wall during hypertension are termed "eutrophic remodeling" [82] and subsequently translate to other vascular pathologies. This involves an inward encroachment of the arterial wall thereby, reducing the diameter of the lumen [83] . Several RAAS components are involved in neointimal formation after vascular endothelial damage [84] . In particular, Rakugi et al. [85] observed that vascular endothelial damage results in the induction of vascular ACE. Their results suggested that inhibition of vascular ACE might be critical in the prevention of restenosis after balloon injury. Patients with previously untreated essential hypertension and eutrophic inward remodeling appears to respond to antihypertensive medication. Reduction in BP with drugs that block the RAAS such as ACEIs [86] [87] [88] or ARBs [86, 87, 89] and calcium channel antagonists [90] are able to reverse the eutrophic inward remodeling [88] . The protein and mRNA of ACE2 are expressed in human coronary arteries and arterioles and the vasa vasorum of most organs [9, 91] . Recently, ACE2 expression has also been (1-7) , Ang-(1-9) and Ang II in DOCA salt hypertensive rats treated with the Rho kinase inhibitor fasudil. Increased plasma levels of Ang-(1-9) were observed in DOCA salt hypertensive rats treated with the Rho kinase inhibitor fasudil. Fasudil (100 mh/kg/day) by gavage was administered during 3 weeks, starting on the third week after DOCA administration. Data are presented as mean ± SEM (n = 8-11/group). DOCA: deoxycorticosterone, F: fasudil. * P < 0.05 compared to both Sham and untreated DOCA groups (adapted with permission from [78] ). observed in the large conduit arteries (aorta and carotid) in the HR [92] . ACE2 localizes preferentially in endothelial cells and arterial smooth muscle cells (SMCs) [9, 91] . As for the role of ACE2 in vascular remodeling, the effect of ACE2 on neointima formation has not yet been studied, but Ang-(1-7) infusion after balloon-catheter injury of the rat carotid artery reduced neointima formation [93] . This effect was probably mediated by its inhibition of vascular SMC proliferation [94] . In hypertensive animal models, ACE2 mRNA and protein were associated with immunoreactive Ang-(1-7) in the large conduit arteries of SHRs. Treatment with an ARB induced a fivefold increase in ACE2 mRNA and was associated with a significant increase in aortic Ang-(1-7) protein expression. This effect was associated with a decrease in aortic medial thickness, suggesting that this may be a protective mechanism in the prevention of cardiovascular events during hypertension [94] . Igase et al. [95] showed that ACE2 protein is expressed not only in the media of the carotid artery but also in the neointima of the ballooninjured carotid artery in SHR. The increase in ACE2 protein expression in the neointima following exposure of the rats to an ARB compared to vehicle was associated with a reduction in neointima thickness. These results lead to the hypothesis that there is a strong correlation between the increase in ACE2 protein in the injured carotid artery of SHR and vascular remodeling during blockade of Ang II receptors [95] . There is known the prothrombotic effect of Ang II [96, 97] and the antithrombotic action of Ang-(1-7) [98] in renovascular hypertensive rats. Thus, in this context, the question arises whether Ang-(1-9) effects are similar to Ang II or to Ang-(1-7) in in vivo conditions. Kramkowski et al. [99] described that Ang-(1-9) enhances electrically stimulated thrombosis in rats and that this effect was abolished by losartan-an antagonist of the AT1 receptor. The prothrombotic activity of Ang-(1-9) was accompanied by the enhancement of ex vivo platelet aggregation and in vitro Ang-(1-9) increased platelet aggregation. However, there are some points in this paper that should be clarified. First, thrombus formation was initiated by electrical stimulation producing arterial injury that is unrelated to a clinical situation. Second, the prothrombotic effect of Ang-(1-9) was much weaker, to the prothrombotic action of Ang II [96, 97] . Third, Ang-(1-9) slightly increased platelet aggregation in in vitro conditions. On the contrary Ocaranza et al. [78] showed that by inhibiting the RhoA/Rho-kinase pathway with fasudil, gene expression and enzymatic ACE activity and plasma levels of Ang II were reduced ( Figure 2 ) and whereas aortic gene expression and ACE2 activity were importantly increased. Simultaneously, plasma levels of Ang-(1-9) (Figure 2 ), mRNA eNOS levels increased and the aortic overexpression of the remodeling promotion proteins TGF-β1, PAI-1, and MCP-1 as well as the increased aortic NADPH oxidase activity and O 2− production were reduced, as a consequence of direct RhoA/Rho-kinase inhibition [100] . This novel effect of RhoA/Rho-kinase inhibition on ACE2 gene expression, enzymatic activity, and Ang-(1-9) levels might additionally contribute to its benefits in hypertension, atherosclerosis, and in cardiovascular and renal pathologic remodeling. This is the first observation concerning a pharmacologic ACE2 and Ang-(1-9) levels activator, both in normotensive and in hypertensive animals, one of the most interesting findings of that study (Figure 2 ). Additionally, in experimental hypertension, direct RhoA/Rho-kinase inhibition also normalizes overexpression of genes that promote vascular remodeling. Interestingly, the observed changes in ACE/ACE2 and in Ang-(1-9) levels were present only during fasudil treatment both in sham and in the DOCA hypertensive rats [78] . Thus, vascular remodeling could be more dependent on the tissue ACE2/Ang-(1-9) axis than on Ang-(1-7) levels in normotensive as well as in hypertensive rats. In vessels, new members of the RAS have been detected, including ACE2, Ang-(1-7) and Mas. Vascular ACE2 is functionally active and generates Ang-(1-7) from Ang II. Ang-(1-7) is found in the endothelium and vascular wall [101] [102] [103] and immunohistochemical staining shows abundant presence in aortic perivascular adventitial tissue [104, 105] . Ang- (1-7) , by binding to receptor Mas on endothelial cells, opposes Ang II actions by mediating vasodilation, growthinhibition, antiinflammatory responses, antiarrhythmogenic and antithrombotic effects [33, 68] through NOS-derived NO production, activation of protein tyrosine phosphatases, reduced MAPK activation and inhibition of NADPH oxidase-derived generation of reactive oxygen species (ROS) [106, 107] . Overexpression of ACE2 in the vascular wall of SHR is associated with improved endothelial function and attenuated development of hypertension [68] . Ang-(1-7)-Mas can hetero-oligomerize with AT1R, thereby inhibiting Ang II actions. The ACE2/Ang-(1-7)-Mas axis is now considered as a counter-regulatory system to the ACE-Ang II-AT1R axis in the vasculature [107] , although some evidence indicates that Ang-(1-7) may also promote fibrosis and inflammation in certain conditions [108, 109] . After myocardial injury or in response to chronically increased hemodynamic load, cardiac mass increases as a result of cardiomyocyte hypertrophy and ventricular wall thickening. Initially these changes are compensatory mechanisms which help to maintain ejection performance and heart function. With continued hemodynamic overload the heart becomes dilated and its walls thinner, resulting in a geometry that contributes to systolic dysfunction by increasing wall stress [110] . At the cellular level, cardiac myocytes increase in size (hypertrophy), rearrange within the myocardial matrix (cell slippage), and die, to be replaced by fibrous tissue, which include fibroblasts and collagen. These changes are collectively referred to as "remodeling" [111] . Cardiac remodeling has been consistently associated with an impaired prognosis in patients with hypertension, MI and chronic heart failure (CHF) [112] . Despite recent advances in our understanding of the ACE2/Ang-(1-7)/axis, the functional role of ACE2 in the heart is somewhat controversial. Crackower et al. [64] originally reported a progressive reduction in LV contractile function in ACE2-null mice without significant changes in fibrosis, left ventricular and cardiac myocyte hypertrophy, or in mean arterial pressure [64] . Interestingly, whereas plasma and tissue levels of Ang II were increased, a decrease in blood pressure was only observed in 6-month-old male ACE−/− homozygote mice but not in age-matched females or 3month-old males. Conversely, Gurley et al. [113] reported that ACE2 deletion enhanced the susceptibility to Ang IIinduced hypertension but had no effect on cardiac structure or function [113] . Huentelman et al. [114] showed that the ACE2 overexpression protects the heart from Ang II-induced hypertrophy and fibrosis. More recently, in SHR hypertensive rats Díez-Freire et al. by using lentiviral-based ACE2 gene transfer, attenuated cardiac fibrosis and hypertrophy [70] and also improved LV and remodeling after experimental MI [115] . Finally, Yamamoto et al. [116] reported that ACE2 deletion exacerbated pressure overload-induced cardiac dysfunction and remodeling that was associated with increased intracardiac Ang II levels and AT1R activation. The reasons for these discrepancies seem to be: (a) the genetic background of the mice used for ACE2 gene deletion [113] , (b) global versus tissue-specific ACE2 manipulation, or (c) the cardiac responses were monitored under basal or pathophysiological conditions. In MI Ocaranza et al. [51] observed that (a) circulating and LV enzymatic activities of ACE2 were downregulated in the long-term phase of LV dysfunction in rats, (b) these effects were prevented by the conventional ACE inhibitor enalapril, (c) plasma Ang-(1-9) levels were significantly increased when MI rats or sham-operated rats were treated with enalapril for 8 weeks but circulating Ang-(1-7) levels did not change at that time ( Figure 1) Ang-(1-9) Figure 3 : Signaling events and cellular effects induced by Ang II via AT1R and opposing effects of Ang-(1-9) acting through AT2R. Proposed Ang-(1-9)-dependent mechanisms that antagonize the cardiovascular remodeling effects of Ang II. ACE2 can directly cleave Ang I to form Ang- (1) (2) (3) (4) (5) (6) (7) (8) (9) . This peptide activates the AT2R to initiate signaling pathways that antagonize AT1R-mediated tyrosine kinase cascades. In this simplified scenario, Ang-(1-9) increases SHP-1 tyrosine phosphatase activity to inactivate src-dependent signaling. AT2R activation also acts other pathways such as NO-AKT. AT1R: Ang II type 1 receptor; AT2R: Ang II type 2 receptor; ERK1/2: extracellular signal-regulated kinase 1/2; JAK: Janus-activated kinase; MAPK: mitogen-activated protein kinase; p38: p38 MAPK; PKC: protein kinase C; STAT: signal transducer and activator of transcription; NO: nitric oxide; SHP-1: protein tyrosine phosphatase SH2 domain-containing phosphatase 1; MEK: mitogen/ERK kinase. Solid arrows indicates activation broken arrows indicates inactivation. findings, it was proposed in this model of HF, that Ang-(1-9) rather than Ang-(1-7) acts as a counterregulator of Ang II [51] . Recently, in MI rats randomized to receive either vehicle, the ACEI enalapril, or the ARB candesartan for 8 weeks, Ocaranza et al. [52] observed that both drugs prevented LVH and increased plasma Ang-(1-9) levels by several folds. Ang-(1-9) levels correlated negatively with different LVH markers with or without adjustment for BP reduction. This effect was specific as neither Ang-(1-7), Ang II nor bradykinins were correlated with LVH. Chronic administration of Ang-(1-9) to MI rats by osmotic minipumps versus vehicle for two weeks decreased plasma Ang II levels, inhibited ACE activity and also prevented cardiac myocyte hypertrophy. Because there are in vitro evidences that the incubation of Ang-(1-9) with ACE generates Ang-(1-7) [9] , and Ang-(1-7) negatively regulates hypertrophy [34, 117] , the authors used the Ang-(1-7) receptor blocker A779 to investigate whether Ang-(1-7) could mediate the effects of Ang-(1-9). Even though A779 was bioactive, with significant increase in circulating Ang-(1-7) levels by 2.7 fold, this compound did not modify the Ang-(1-9)-dependent suppression of cardiac myocytes hypertrophy induced by MI [52] . In in vitro experiments with cardiac myocytes incubated with norepinephrine (10 μM) or with IGF-1 (10 nM), Ang-(1-9) also prevented hypertrophy and this effect was not modified by the coincubation with Ang-(1-9) and A779 [52] . To further understand the role of Ang-(1-9) compared to Ang-(1-7) in cardiomyocyte hypertrophy, Flores-Muñoz et al. [56] studied Ang-(1-9) effects in rat neonatal H9c2 and in rabbit left ventricular cardiomyocytes. Cardiomyocyte hypertrophy was stimulated with Ang II or vasopressin, significantly increasing cell size by approximately 1.2-fold as well as stimulating expression of the hypertrophy gene markers atrial natriuretic peptide, brain natriuretic peptide, β-myosin heavy chain and myosin light chain (2-to 5fold). Both Ang-(1-9) and Ang-(1-7) were able to block hypertrophy induced by either agonist. The effects of Ang-1-9) were not inhibited by captopril, supporting previous evidence that Ang-(1-9) acts independently of Ang-(1-7). The authors investigated receptor signalling via angiotensin type 1 and type 2 receptors (AT1R, AT2R) and Mas. The AT1R antagonist losartan blocked Ang II-induced, but not vasopressin-induced, hypertrophy. Losartan did not block the antihypertrophic effects of Ang-(1-9), or Ang-(1-7) on vasopressin-stimulated cardiomyocytes. The Mas antagonist A779 efficiently blocked the antihypertrophic effects of Ang-(1-7), without affecting Ang-(1-9). Furthermore, Ang-(1-7) activity was also inhibited in the presence of the bradykinin type 2 receptor antagonist HOE140, without affecting Ang-(1-9). Moreover, Flores-Muñoz et al. [56] observed that the AT2R antagonist PD123,319 abolished the antihypertrophic effects of Ang-(1-9), without affecting Ang-(1-7), suggesting Ang-(1-9) signals via the AT2R. Radioligand binding assays 8 International Journal of Hypertension demonstrated that Ang-(1-9) was able to bind the AT2R (pKi = 6.28 ± 0.1). The data indicate that ACE2/Ang-(1-9) axis, acting as a counterregulator of Ang II, is an effective, and possibly direct novel anticardiac hypertrophy axis. Pharmacological treatments based on the RAS blockade are used extensively for the treatment of hypertension and CV remodeling. However, in spite of their success in pharmacological blockade of the RAS, the prevalence of end-organ damage has risen steadily in the last several decades. These observations indicate that novel and innovative approaches must be used in an attempt to promote a more effective treatment for the residual CV remodeling. In this environment, the ACE2/Ang-(1-9) axis is an important target, that is critical in tipping the balance of vasoconstrictive/proliferative to vasodilatory/antiproliferative axis of the RAS. Conceptually, the ACE2/Ang-(1-9)/AT2 axis balances the adverse effects of the ACE-Ang II-AT1 receptor axis ( Figure 3 ). Accumulating evidence suggests that ACE2 expression and Ang-(1-9) levels are altered in diastolic and systolic dysfunction and remodeling and the activation of the ACE2/Ang-(1-9) axis protects the heart and vessels from cardiovascular remodeling. In conclusion, the noncanonical RAS arm has new biological effector Ang-(1-9) to counterregulate the classical RAS.

Viral Proteins Acquired from a Host Converge to Simplified Domain Architectures

The infection cycle of viruses creates many opportunities for the exchange of genetic material with the host. Many viruses integrate their sequences into the genome of their host for replication. These processes may lead to the virus acquisition of host sequences. Such sequences are prone to accumulation of mutations and deletions. However, in rare instances, sequences acquired from a host become beneficial for the virus. We searched for unexpected sequence similarity among the 900,000 viral proteins and all proteins from cellular organisms. Here, we focus on viruses that infect metazoa. The high-conservation analysis yielded 187 instances of highly similar viral-host sequences. Only a small number of them represent viruses that hijacked host sequences. The low-conservation sequence analysis utilizes the Pfam family collection. About 5% of the 12,000 statistical models archived in Pfam are composed of viral-metazoan proteins. In about half of Pfam families, we provide indirect support for the directionality from the host to the virus. The other families are either wrongly annotated or reflect an extensive sequence exchange between the viruses and their hosts. In about 75% of cross-taxa Pfam families, the viral proteins are significantly shorter than their metazoan counterparts. The tendency for shorter viral proteins relative to their related host proteins accounts for the acquisition of only a fragment of the host gene, the elimination of an internal domain and shortening of the linkers between domains. We conclude that, along viral evolution, the host-originated sequences accommodate simplified domain compositions. We postulate that the trimmed proteins act by interfering with the fundamental function of the host including intracellular signaling, post-translational modification, protein-protein interaction networks and cellular trafficking. We compiled a collection of hijacked protein sequences. These sequences are attractive targets for manipulation of viral infection.

Many studies, mainly from bacteria and unicellular eukaryotes, focus on the exchange of genetic material between viruses and cellular hosts. Sequences are best studied through their structural and functional domains [1, 2, 3, 4, 5] . The evolution of domains is a significant force for shaping the proteins along the tree of life. Sequence exchange between genomes within and between superkingdoms is evident from the appearance of a domain in a particular phylogenetic branch [6] . The contribution of horizontal gene transfer is not limited to bacteria but has occurred across distant species [3] . For example, some signaling domains in bacteria are the consequence of a horizontal gene transfer [7] . The viruses are parasitic agents that maintain an intimacy with their host cells. Consequently, an extensive horizontal evolution [8] is associated with the viral life cycle. The lack of similarity of viral proteins (e.g., capsid proteins) with any cellular organisms is in accord with their early and unique origin [8, 9] . Most likely, the modern viruses originated at the early RNA world of the primordial genetic pool. With the increasing numbers of sequenced viruses, similarity among seemingly unrelated viruses was reported. A role of the hosts as vehicles for such cases is proposed. For example, the structural similarities observed between bacterial viruses (PRD1, Bam35), Chlorella virus (PBCV-1) and adenovirus in the coat proteins, led to the proposal that all viruses are old, probably preceding the cellular life. Furthermore, it is compatible with polyphyletic virus origins, as opposed to the monophyletic origin of cellular life [10] . Still, assignment of viruses to the phylogenetic tree of life remains unresolved [11] . Notably, viruses as vectors (mainly RNA viruses) have the potential to rearrange the genomic material, and thus, to change the domain architecture [12, 13, 14] . Studies on horizontal gene transfer focused primarily on viruses infecting bacteria and archaea (e.g., bacteriophages) [15, 16] . The co-evolution of viruses toward their hosts indicates an active crosstalk on an evolutionary time scale [17, 18, 19] . Several studies reported on a handful of cases of functional mimicry by viral proteins [20] . In few cases, evidence for gene transfer from the host to the virus is obvious. For example, the photosynthetic efficiency in cyanobacteria (Synechococcus and Prochlorococcus) relies on components of the photosystem II. These critical components express in the respective phages [21] . In the case of the phytoplankton-virus system, the DNA virus EhV that infects the microalge (Emiliania huxleyi), contains a complete metabolic pathway as a result of a horizontal gene transfer [22] . A similar case is demonstrated for the dUTPase genes (Dut) that are necessary for regulating the cellular levels of dUTP. Phylogenetic analysis revealed the origin of the viral Dut sequence in a monophyletic cluster of DNA viruses with eukaryotic hosts [23] . The Acanthamoeba polyphaga Mimivirus and the family Phycodnaviridae [24] , contain many genes that are found in cellular organisms. For example, the giant virus Cafeteria roenbergensis virus (CroV) includes numerous eukaryotic-like genes for translation factors, ubiquitin pathway components, intein elements, histone acetyltransferase and more [25] . These are extremely large viruses of aqueous environments that infect bacteria, animals and protists [26] . A search for similarities between viral and host proteins has largely been focused on herpesviruses [27] , Hepadnaviridae [28] and others. However, the high mutation rate of RNA viruses [29] and the coexistence between viruses and their hosts for millions of years has most likely blurred the sequence similarity. Recently, several studies challenged the origin of ancient viral segments in metazoan genomes. These sequences that are called EVE (for endogenous viral element) encompass all virus-derived genomic loci [30] . In this paper, we present a coherent survey on protein sequences that are shared between viruses and their hosts. We assess the scale of the phenomenon by focusing on the viral-related protein sequences that appear in metazoa. We have used the current archive of all proteins [31] as the basis for identifying sequences with a potentially common origin. Presumably, their appearance in the virus reflects virus-acquired sequences. Of about 190 instances of highly similar viral-eukaryotes sequences, we recognize that only a small number originated from a host origin. We extended the collection of viral proteins that have a host origin by investigating the eukaryotes-viruses Pfam families [32] . We focused on the 670 Pfam cross-taxa families that contain viruses and metazoa. A careful examination reveals that these instances reflect either missed annotations or the remnants of sequence exchange by virus infection. To distinguish these possibilities, we constructed sequence alignment trees for all 670 Pfam families. From the properties of the trees, we focused on 335 families that most likely contain viruses that hijacked sequences from their host. We found that most of the viral proteins in the orthologous families are much shorter and composed of simpler domain architectures. In almost all cases, the number of domains and the sequence of the tails and the inter-domain linkers are considerably shorter in the viral proteins relative to their counterpart host proteins. We discuss the potential of such short viral proteins to interfere with critical cellular functions and thus are candidates for manipulation strategies in defeating viral infection. Genetic material exchange between viruses and their hosts Figure 1A shows two over-simplified scenarios in support of a genetic exchange from the virus to the host genome and in the reverse direction, from the host to the viral genome. In the first scenario, a viral sequence is detected in the host (e.g., human) but not in the rest of the phylogenetic branch. The following scenario accounts for viral sequences acquired from the host ( Figure 1A , right). Under this scenario, the viral gene sequence is identified in a broad group of organisms that belong to a phylogenetic tree that includes the host (human). Therefore, the sequence in the virus is most likely a reflection of a hijacking event, according to an argument of maximum parsimony. Supporting evidence for the directionality of the genetic exchange of viral and cellular organisms relies on a detailed phylogenetic analysis. The topology of the reconstructed tree is used to support the most parsimonious scenario (see Materials and Methods). The simplified illustrations in Figure 1A do not address the more complicated, realistic instances in which different viruses carry sequences that resemble various organisms. An additional criterion used in supporting the occurrence of sequence acquisition by viruses is the presence of a sequence resemblance in the known host. The origin of viruses is probably preceding the cellular life [8, 10] . Thus, the ancient events in which viral sequences were incorporated into an ancestor eukaryote cannot be traced by their sequence similarity. Still, a conserved functional or structural similarity could expose such early events [33] . In this study, we have not attempted to date the horizontal transfer event. Furthermore, we will not discuss the events of genetic material exchange (see discussion in [34] ), but limit our study to the acquisition of coding sequences in viruses and metazoa. There are about one million viral proteins in the UniProt database (990,049, August 2010) that represent about 66,000 viral strains. This is a highly redundant resource and about half of it composed of medically relevant strains including Hepatitis B viruses (HBV) and Human immunodeficiency virus (HIV). We took advantage of a reliable source of UniRef [35] that unifies sequences according to their identity level along the sequence length. We used UniRef90 classification (see Materials and Methods). There are .165,000 UniRef90 clusters that contain at least one viral protein ( Figure 1B) . However, from this set, we only considered 262 instances that contain at least two proteins, where one of them must be a eukaryote ( Figure 1B) . Of the 5,482 cross-taxa clusters that contain sequences from viruses and cellular organisms, 95% are sequences of bacteriophages and plasmids confined to the bacteria [36] . We will not further discuss the events that are confined to bacteria and archaea. A taxonomical view shows the diversity of the organisms that share the UniRef90 clusters with viral proteins ( Figure 1C ). It shows that the eukaryotes are the most diverse group with 106 species that share their homologues with viral proteins. This result Many studies focused on the exchange of genetic material between viruses and cellular hosts. The diversity of viruses argues that, along the evolutionary history, viruses have shaped the host genomes. While most viruses have many opportunities to exchange genetic material with their hosts, tracing such events is challenging as the origin of the sequences is masked by the high mutation rate of many viruses. On the other end, for completing a successful infection cycle the viruses must cope with the cell machinery for entry, replication and translation while hiding from the host immune system. We collected evidence for instances of viral protein sequences that were most probably ''stolen'' from the hosts. Additionally, a shared ancestry with metazoa is associated with 670 Pfam domain families. For half of these families, the origin of the viral proteins from its host is supported. For about 75% of the cross virus-metazoa families, the viral proteins are significantly shorter than their counterpart host proteins. Most of these cross-taxa viral proteins are single domain proteins and proteins with a simple domain composition relative to the proteins of their hosts. These viral proteins provide insights on the overlooked intimacy of viruses and their multicellular hosts. suggests that the phenomenon of shared sequences is quite broad, and many eukaryotes have been subjected to a genetic material exchange. Among the UniRef90 clusters that contain viruses and eukaryotes ( Figure 1B) , ,70% are from tetrapoda, 13% plants, 13% arthropoda, 4% fungi and only a smaller percentage of other taxa. We focused on the cross-taxa clusters of viruses and mammals (118 clusters include ,2,200 proteins, Figure 1D ). Viruses from Class I (dsDNA viruses with no RNA stage) and class VI (Retro-transcribing ssRNA, Plus strand) are prevalent among those that infect mammals [17] . The dominating Class VI viruses are characterized by their ability to integrate sequences into the host ( Figure 1A , left). Table 1 lists the Class I viral proteins that share sequences with mammals (23 clusters, Figure 1D ). In Class I viruses, the virus enters the nucleus before its replication (with the exception of Poxvirus family) and its infectivity is strongly dependent on the host cell division. Discriminating whether a sequence has originated from the virus or the host is not straightforward. We generated for each cluster a phylogenetic dendogram and analyzed the connectivity of the viral protein in view of its neighboring sequences. Often the analyzed cluster is too small. In such cases, we expanded the cluster to the relaxed UniRef50 classification. We applied additional criteria in support of a virus having acquired protein sequences from the host: (i) The tested sequence appears in several organisms ($2) on the same evolutionary branch (as in Figure 1A , right); (ii) The tested sequence is not associated with viral contamination. Most analyzed cross-taxa clusters derive from the contamination by viral proteins following integration of the virus to a mammalian genome. Several instances were contaminated by the extensive use of viral vectors as vehicles in variety of molecular manipulations (e.g., Adenoviruses, Table 1 ). Another source of contamination is from cancerous cells infected by viruses (e.g., human papilloma virus). In such instances, some sequences that are assigned as 'human' are incorrectly annotated. In these instances that reflect the incorporation of the virus to the host, different protein sequences from the same virus are identified which are best explained as a result of infection or an integration event. For example, the proteins in UniRef90 clusters P06426, P06463, P21735, P06788, P36741 and P21736 belong to Human papilloma virus (Table 1) . Studies on the viral sequences that were integrated into the vertebrate germ line and hence shaped the vertebrate genetic heritage were reported [37, 38, 39] . Herein, we only consider the protein sequences that are shared by viruses and their metazoan hosts. The principal virus families that infect multicellular eukaryotes are listed in Supportive data Table S1 . For few instances, a support exists for viruses that hijacked sequences from the host. Among Class I viruses (Table 1 ) the shared functions include interlukin-10 (IL-10) (Figure 2 ), beta-1,6-Nacetylglucosaminyltransferase (b1,6GnT) ( Figure S1 ) and Ubiquitin. The b1,6GnT and IL-10 are found exclusively in metazoa and the multicellular eukaryotic branch. The key features and the functional amino acids are conserved in the viral and the corresponding mammalian proteins (Figure 2A , Figure S1 ). Indeed, in human cells lacking b1,6GnT gene, the Bovine herpesvirus 4 (BoHV-4) sequence fully recovered the missing enzymatic activity [40] . Resolving the evolution of the Ubiquitin in the genome of Pestivirus suggested that the virus hijacked Ubiquitin-related sequences in two consecutive events [41] . A browsable table is available at www.protonet.cs.huji.ac.il/ virost/tables/UniRef90-Class1.html. Figure 2 shows a prototypic case of viral proteins that resemble the host protein. Interleukin 10 (IL-10) inhibits the induction of pro-inflammatory cytokines. IL-10 was found in many viruses including Epstein-Barr virus (EBV), equine herpesvirus (EHV) and cytomegalovirus (CMV) [42] . Presumably, the gene product protects the infected cells from the host defense mechanism. An extended cluster of IL-10 (Table 1) covers 20 viruses and 96 cellular organisms (UniRef50_P22301). Representatives of viral and metazoan proteins are shown by the multiple sequence alignment (MSA) ( Figure 2B ). Most of the variations in the viral and metazoan protein reside in the sequence of the N-terminal that covers the signal peptide ( Figure 2B ). Traces of a genomic organization of the host in the viral genome were reported. For example, IL-10 like sequence from the gammaherpesvirus ovine herpesvirus 2 includes 5 exons and 4 introns [43] . Inspecting the UniRef90 clusters that contain proteins from viruses and metazoa (187 clusters) shows a wide variation in the distribution of protein lengths (Supportive data Table S2 ). Viruses tend to reduce their production load by deleting and reducing the unessential genetic material [44] . While this length reduction is an absolute necessity for most viruses, some giant viruses (e.g., Mimivirus, Chlorovirus, and Cafeteria roenbergensis virus) include ,1000 proteins [24, 25] . The evolution origin of proteins from the Giant viruses remains unknown [45] . Still, 12% of the Acanthamoeba polyphaga mimivirus (APMV) proteins constitute a large number of host related sequences. The average length of this subset of the Mimivirus proteins (523 amino acids) is similar to the length of their homologous sequences. The small numbers of cases of viral acquired sequences (Table 1, Supportive data Table S2 ) may indicate the sequence divergence that had occurred throughout evolution. We therefore expanded the analysis for remote homologous. We questioned whether the viral protein sequences that were already substantially diverged due to a rapid evolution rate, or a long evolutionary history still maintain the host protein's functional domain. The Pfam provides a comprehensive resource of functional and structural families and domains. Each Pfam entry represents a statistical model with an average sequence identity of 30-40% among the members of the family. Currently, Pfam covers 11,912 families, where 1,165 families include at least a viral protein and a eukaryotic protein representative. Some Pfam families are extremely large. Among families that contain metazoa and viral proteins are 'Helix-loop-helix DNA-binding domain' (,6000 proteins) and 'Sugar transporter' (,12,000 proteins). Contamination of viral proteins in metazoan proteomes (e.g., Capsid, Env, Tat) occurs mainly as a result of viral vector manipulations in cell lines, leading to incorrect assignment as a viral-eukaryotic crosstaxa family. An example is the GFP family (PF01353) that we have manually removed from the analysis. To reduce such sporadic instances, we considered Pfam families having at least two metazoan proteins, resulting in a list of 667 Pfam families. Supportive data Table S3 lists the species, composition of the domains and the proteins' length. The relatively small numbers of cases of viral acquired sequences (Table 1, Supportive data Table S2 ) may indicate the sequence divergence that had occurred throughout evolution. Therefore, we expanded the analysis for remote homologous. We questioned whether the viral protein sequences that were already substantially diverged due to a rapid evolution rate, or a long evolutionary history still maintain the host protein's functional domain. Over 300 cross-taxa Pfam families (virus-metazoa) are best explained by a viral acquisition of host sequences. Instances of lateral gene transfer between bacteria and their bacteriophages dominate many of the cross-taxa Pfam families. Other families contain genuine viral proteins contaminated by metazoan proteins. In order to justify the directionality of sequences from the hosts to the virus, we constructed for each of the 667 Pfam families a sequence-based tree (MSA based on the domain and not the full length sequence). We considered Pfam families in which only 1-2 viral proteins are included in the family, and families in which the percentage of the virus proteins in the family is small (,5%, Figure 3A ). The vast majority (547 families, 82%) of the analyzed Pfam families fulfilled these criteria ( Figure 3A , blue). We also requested that the viral proteins are clustered in sub-trees within the family tree. We counted the number of viral proteins spreading within the sequence alignment tree. We suggest that viral proteins that are clustered in a defined sub-tree (called Viral Cluster, VC) are likely to represent a single episode of acquired sequence from the host. Consequently, only a limited diversity among the closely related viruses is expected in view of the rest of the tree. 64% of the 547 Pfam families from the previous selection fulfill the requirement for clustered viral proteins. These are the Pfam families that contain #2 viral clusters (60%), and other families (4%) that are specified by a high degree of condensation (i.e. the ratio of the viral proteins to the number of VCs is $3). These filtration steps further reduced the list of relevant Pfam families to 335 ( Figure 3B ). We show a tree constructed for one of the 335 families. The IL-6 (PF00489) family contains 10 viral proteins ( Figure 3C , blue) that are split to two sub-trees of viral clusters (VC) and other 136 Metazoan proteins (marked as collapsed sub-trees). The maximal depth in this tree is 19 (included in the collapsed sub-tree, red triangle). The deepest viral protein in the tree is of depth = 9, and its normalized depth is 9/19 = 0.474. The depth of the viral cluster (VC, the maximal sub-tree which contains all viral proteins) is therefore, 4/19 = 0.211. The normalized depth for all the proteins in the 335 Pfam families ( Figure 3D, top) is analyzed in view of the distribution of the normalized depth of the viral proteins within these families ( Figure 3D, bottom) . It seems that the two distributions are remarkably different ( Figure 3D ) which is in accord with the notion that the viral proteins are relatively isolated subsets among the proteins from the cellular organisms in the relevant Pfam families. Table 2 shows a sample of these families along with the cellular process and the protein function in the viral life cycle. A full list of the 667 Pfam families with the analyzed properties of their alignment trees is provided in Supportive data Table S3 . One of the families that exemplified the trend found in virusmetazoa Pfam families is the PAAD/DAPIN/Pyrin family (PAAD_DAPIN, PF02758). This domain family is a diverse family (26% average sequence identity) that includes 34 cellular species and 5 dsDNA viruses that belong to the Poxviridae. The PAAD domain is at the N-terminal regions of proteins. This domain occurs in several multicellular organisms, in the context of inflammation, signaling and apoptosis ( Figure 4) . Several observations could be extracted for the PAAD domain: (i) Based on a multiple sequence alignment (MSA) of the PAAD domain sequences it is evident that the 5 viral proteins were diverged significantly ( Figure 4B ). All 5 viral proteins reside in one cluster, in the phylogenetic tree, together with other mammals as their sibling in the tree ( Figure 4A, blue font) . The domain architecture within the protein of the family is best explained by an initial extensive duplication of the PAAD domain ( Figure 4A Figure 4A , red font). Note that these proteins spread throughout the sequence-based tree. Presumably, it is a reflection of a domain loss event. Some of these proteins are fragments (e.g., Q5T3V8_HUMAN), and others include less characterized PfamB domains [32] (e.g., IFI4L_MOUSE, Q3UPZ5_MOUSE). The initial tests on UniRef90 covered 14,000 proteins in relatively small clusters (,90 proteins on average, Supportive data Table S2 ). In contrast, the collection of the cross-taxa Pfam families (Table S3) covers 161,000 viral proteins and 400,000 metazoan proteins. Therefore, focusing on the cross-taxa Pfam families provides an opportunity to increase the statistical power of the tests. Several statistical observations regarding the sequences among the cross-taxa families of viruses and multicellular organisms can be made: (i) The average length of the metazoan proteins is 507 amino acids, while the average length for the viral proteins in these families is only 396 amino acids (P-value of ,1.0e-17 by the KS-test, Figure 5A ). (ii) For 73% of all families, the viral proteins are shorter than the length of the average metazoan proteins in the family (P-value,1.0e-13 by the Hypergeometric test). (iii) In 67% of the families, the number of Pfam domain appearances (including several repeats of the same domain or different ones, Figure 5B ) is smaller in the viral proteins relative to the metazoan proteins in the family (P-value,1.0e-40 by the KS test). (iv) In 62% of the families, the number of different Pfam domains is higher in the metazoan proteins relative to the viral proteins. (v) For the discussed families, the median number of Pfam domains is 1.06 while, for the metazoan proteins, this value is 1.7 (Pvalue,1.0e-32 by KS test). Many metazoan proteins are multi-domain (colored rectangle, Figure 5B ). We tested whether the viral acquired sequences that belong to multi-domain proteins displayed a stronger tendency for a size reduction (see scheme, Figure 5B ). A reduction in length of viral proteins may be a reflection of reducing the number of domains ( Figure 5B, b-c) , shortening the length of the linker sequences ( Figure 5B , a) or even the trimming of the length of the domain itself. Among the 667 analyzed Pfam families, in 103 of them, the metazoan proteins contain at least 3 Pfam domains. In 85% of this set (88 families), the viral proteins are shorter ( Figure 5B, virus) . Remarkably, the average length of these 103 metazoan proteins families is 912 amino acids relative to 503 amino acids for the viral proteins that belong to these families. Similarly, in this set of multidomain proteins the viral proteins have an average of 2.9 domains, while the metazoan proteins have 4.6 domains on average (paired t-test, p-value of 1.0e-11). This shows that the tendency to reduce the protein length and the number of domains is stronger when the number of Pfam occurrence in the original host protein is higher. In order to reduce the risk of misclassification, we further restricted the analysis to Pfam families of viruses-metazoa (with $3 Pfam domains) that contain at least 2 viral proteins (total of 50 families). The length of the viral proteins is significantly reduced. For 90% of these families (above the reference line, Figure 5C ), the viral proteins are shorter than their matched metazoan proteins. In order to determine whether the reduction in length is due to a reduction in the number or the properties of the domains, we repeated the analysis for the ratio of the number of distinct domains (depicted by the different colored rectangles, Figure 5B) in the viral and their relevant metazoan sequences ( Figure 5D ). For 80% of the families (families above the reference line), number of different Pfam domains that are associated with viral proteins is reduced. Note that by this measure, a short viral protein ( Figure 5B a-b) still has a ratio of 1.0. We show that the viral proteins are not only significantly shortened, but have also converged to a simpler domain composition. The length of the individual domains between the viral and the metazoan host proteins is identical (Supportive data Figure S2 ). Recall, that this observation may be mainly due to the definition of belonging to a Pfam domain family. The high statistical significance of these trends is consistent with a possibility that the short viral proteins have resulted from the acquisition of fragments from the host protein. Alternatively, it can be the result of a refinement of the acquired sequences during viral evolution. We separated each protein into three segments: (i) The Pfam domain(s); (ii) The tail linker (TAIL) that combines the amino acid extension towards the N-and the C-termini of the protein, beyond the boundary of the domain(s); (iii) The internal domain linker (IDOL) that comprises the sum of the amino acid spacers between domains. Clearly a single domain protein lacks IDOL. We performed a separate analysis for the TAIL and the IDOL sequences ( Figures 6A-6D) . The study was performed on all the families that have at least 2 Pfam domains (unique or repeated). The average TAIL in viral proteins is 14 amino acids while the metazoan protein TAIL length is 85 amino acids (p-value,1.0e-150, Figures 6A-6B ). Trimming of protein tails at both termini often leads to a loss of cellular localization signals (e.g., KDEL, PDZ binding sites are found at C-termini) [17] . Importantly, the average IDOL length of the viral proteins in the Pfam families is 30 amino acids, while, for the metazoan equivalent proteins, the length is 67 amino acids (p-value,1.0e-150, Figures 6C-6D) . While a short TAIL may be explained by the viruses having acquired a fragmented sequence from their hosts, the same trend was found for the IDOL. Figure 6C shows that while only 54% of the metazoa IDOL have a length of ,40 amino acids, in the viral proteins from the same Pfam families, 96% of the proteins have IDOL that are shorter than 40 amino acids. These results are consistent with an active trimming and refinement process throughout viral evolution. Short IDOL length is advantageous in suppressing protein misfolding, and hence, improving translation effectiveness [46] . Similarly to the finding of short IDOL sequences in viral proteins, we identified instances of an internal domain which is missing in the viral protein while the flanking domains are maintained in the same order in the eukaryotic homologous protein ( Figure 6E) . The viral putative phosphatidylinositol kinase L615 (UniProt: Q5UR69) is a 701 amino acid protein from the Acanthamoeba polyphaga Mimivirus (APMV) that infects Amoeba. It has two Pfam domains: FYVE (PF01363) followed by PI3_PI4_kinase (PF00454). There are no other known proteins with identical domain architecture in the Amoebozoa kingdom (there are 7 such proteins in other kingdom, e.g., Stramenopiles and Excavata). However, there are 3 proteins from the genus Dictyosteliida (slime molds) that do belong to the Amoebozoa kingdom (UniProt D3BQ22, Q54UU9 and EGC34678). In all 3 of these proteins, the architecture is composed of FYVE domain followed by PI3Ka and PI3_PI4_kinase. The missing domain of PI3Ka in the Mimivirus (APMV) provides an evidence for an active elimination of an internal domain based on parsimonious argument. The findings of shorter IDOL (Figures 6C-6D) or absence of internal domains ( Figure 6E ) are probably the result of the trimming and shortening of the sequences after their acquisition by the virus. The possibility of a domain insertion in eukaryotes cannot be excluded. The exhaustive search for sequences that were hijacked by viruses from their host allowed us to speculate on the underlying modes of mimicry. It was shown that once a mimicry function by a virus is established, the corresponding functional partner protein of the host undergoes a fast positive selection to overcome the deleterious effect of the viral mimicry [20] . According to these findings, the viral proteins that originated from the hosts are short versions of the full-length host proteins ( Figures 5-6, Supplemental Figures S2, S4) . Furthermore, these proteins are characterized by a substantial reduction in the architectures of the domains ( Figure 5 ) and the protein linkers ( Figure 6 ). We classified these proteins into distinct (yet not exclusive) modes of action. For simplicity, we unified the viral acquired sequences from the cross-taxa families to 5 strategy modes (Figure 7) . Mode A depicts a competition on a receptor binding by a viral ligand that replaces the natural one. Examples for this mode are the expression of the secreted IL-10 ( Figure 2 ), IL-8 (UniProt: Q98158, Q98314, D2E2Z5) and PDGF (UniProtKB Q80GE8, Q2F842 and D0VXD7). These secreted mitogens are identified in class I and class VI viruses (Table 2) . Viral proteins participate in a rich protein-protein interaction (PPI) network [47] . Mode B illustrates PPI, where the virus uses an acquired sequence for replacing a host partner protein or for interacting with a preexisting protein complex. The result is an alteration of the cells' function. Examples for viral proteins that interfere with the host PPI are the anti-apoptotic Bcl-2 sequences and Profilin (Table 2, for example UniProt: Q5IXM3, P33828, P68695) . Mammalian Semaphorins (Sema7) and the Smallpox virus A39R protein ( Table 2 , UniProt: Q775N9, B7SV99, Q0N658, A0ES13) share identical binding modes with a crossreactivity towards common receptors [48] . Mode C depicts the role of protein modifications (e.g., phosphorylation). A viral protein can either mimic the host modifications (Figure 7, marked C1) . Alternatively, a modification occurs by a viral enzyme (Figure 7 , marked C3). Such mimicry can lead to a modification of the original site or at an entirely new site (Figure 7, marked C2) . Apparently, there are instances in which both the modifying enzyme and the target proteins are both sequences that were acquired from the host (Figure 7, marked C4) . This mode is dependent on the presence of active kinases (or phosphatases). For example, human cytomegalovirus (HCMV) kinase introduces phosphorylation sites that perfectly mimic the function of the cellular CDK2 (cyclin dependent kinase) [49] . An evolutionary tree alignment for viral B1R protein kinase (Supplemental Figure S3 ) supports the functional overlap and mimicry with the closely related cellular kinases. Mode D depicts the importance of nucleic acid regulation of transcription. In this mode, a viral protein mimics the host regulation by either competing for an existing transcription factor (Figure 7, marked D1) , or by modifying the transcription program following a DNA/RNA binding (Figure 7 , marked D2). For example, the Epstein-Barr virus (EBV) encodes an activator protein that is similar to Fos/Jun family (bZIP_1, PF00170. For example, UniProt: Q80GR6, Q8QQX9, Q6USE5, D2Y5S7). The difference in specificity and the dimerization properties of the EBV activator allows the activation of an alternative transcription program [50] . Mode E collectively points to the generic strategies for damaging and deactivating the host proteins. It could be achieved by protein tagging (i.e., SUMO, ubiquitin), or the activation of viral proteases. Among the cross-taxa Pfam families, some families are associated with specialized proteases (Table S3) . Mode E shows the various routes by which acquired sequences alter key cellular processes. Molecular mimicry in trafficking and the subcellular localization is common to many viruses. For example, Soluble N-ethylmaleimide sensitive factor Attachment Protein (a-SNAP) is a conserved protein among all eukaryotes. It was also found in Canarypox and Fowlpox viruses [51] . These proteins may alter the balance of the vesicular trafficking, docking and the membrane fusion machinery. In autophagy, viral proteins exploit processes such as membrane fusion and protein folding for the benefit of their replication [52] . We limit the discussion to the modes by which the shorter versions of the viral acquired proteins exhibit their impact on some cellular functions. The described modes (A-E) are effective in additional instances of molecular and functional mimicry [53, 54] . Inspecting the viral proteome is challenging, as the majority of viral sequences are redundant and poorly annotated. Importantly, the rapid evolution and the high mutation rate in some viral classes often leads to the loss of a detectable sequence similarity and, therefore, additional cases of virus hijacking events cannot be detected based on sequence similarity search methods. Despite these drawbacks, we have traced hundreds of viral proteins with respect to their hosts. Only a small fraction of them shows high sequence similarity with corresponding host proteins. For the majority of the cases, the origin of the viral sequences and possible derivations from the host call for applying powerful models for remote homologues. We provided analysis for 670 homologous families (according to the Pfam definition). For half of these families we provided support for sequence acquisition by the viruses from their hosts. The candidate sequences for a host to viral acquisition are useful in exploring the mechanisms by which viruses hijack and refine sequences. We found that most of the viral proteins that potentially originated from host sequences are significantly shorter and contain fewer domains. Furthermore, we propose that the sequence refinement by the virus is a dynamic process. The inter-domain linkers (e.g., sequences connecting domains, but excluding the amino-and carboxyl tails) are significantly short, relative to other related proteins ( Figure 6 ). The viral proteins act in the cell according to a finite number of strategies. The simpler domain composition of these viral proteins is sufficient for the utilization of functional mimicry. Currently, we are expanding the analysis by identifying short peptides in viral proteomes that serve as competition agents for neutralizing critical cellular functions. The collections of 187 UniRef90 clusters and the 667 Pfam cross-taxa families are available as interactive tables. These tables are available at: www.protonet.cs.huji.ac.il/virost/tables/UniRef90.html www.protonet.cs.huji.ac.il/virost/tables/Pfam.html UniProKB includes 990,049 sequences (taxonomy-viruses). The viral proteins include ,15,000 reviewed proteins (UniProt/ SwissProt). The rest of the proteins are from UniProt/TrEMBL. There are 430.6 K sequences after removal of HIV and HBV sequences. Only 241.8 K are full-length (56.1%), while the rest are denoted as 'fragments'. The percentage of full-length proteins in metazoa is 54% (1.191 M/2.2051 M). The pre-calculated classifications of UniRef90 (i.e., identity of .90% at the amino acid level) reduce the UniProKB set to 175,236 clusters. Additional steps of filtrations are: (i) Considering only clusters with a minimal size of 2 proteins (62,129 clusters); (ii) Clusters that also include the metazoan proteins (187 clusters). ViralZone is a database that manually assigns host-virus pairs (http://www.expasy.ch/viralzone, coordinated by UniProt/ SwissProt). ViralZone holds reference strains viruses that belong to 83 families and 330 genera. This is a high quality collection of 'complete proteome'. All viruses are classified into Table S1 . Pfam 24.0 (11,912 families) [32] is a high quality resource for domains and families. A valid cross-taxa list was generated. Eukaryotes and viruses cross-taxa resulted in 1,165 Pfam entries. The following filtration steps were applied: (i) Pfam families with at least one viral protein and at least one metazoan protein (taxid: 33208), total of 859 Pfam families. (ii) Restricting the Pfam to families that have at least one metazoan protein and at least one metazoan-infecting virus resulted in 796 Pfam families. (iii) Pfam families with .95% viral proteins for structural element of the virus (e.g., Env, Coat, Capsid). (iv) Enzymes of the replication system were excluded, as these genes are the outcome of several events of genetic exchange [55] . Specifically, we excluded families of RNA/DNA polymerases (39 families), Exo/ Endonuclease (16 families), Helicase (15 families), tRNA synthetase (8 families) and Primase (8 families). We also manually eliminated the cluster represented by the GFP (PF01353) that reflects the inevitable contamination from the extensive use of GFP as vectors in many molecular biology techniques. The filtered list includes 667 protein Pfam families (Supplemental data Table S3 ). We define linker sequences as TAILs (Tail Linkers) and IDOLs (Inter Domain Linkers). The TAILs are all sequences at the two terminals external to the first and last domain in the protein. Each protein provides two entries. The IDOL is a collection of all interdomain sequences (excluding TAIL). Protein TAIL's length was defined as the mean of the two tail segments. In the same way, IDOL length was defined as the mean of the lengths of the inter domains linkers. We collected the Pfam data for all proteins having at least 2 domains (i.e., having at least one IDOL) and one of the domains belong to the 667 Pfam domain families (Table S3 ). There are ,57,000 such viral proteins and ,98,000 metazoan proteins. Statistical tests were applied for the set of viral proteins in view of the host cellular protein for each cluster (or Pfam family collection). We applied statistical confidence tests (P-values) based on the non-parametric Kolmogorov-Smirnov (KS), Student t-test and the hypergeometric distribution tests. The KS test is based on the maximum distance between the two cumulative curves based on the separated viral and host proteins and viral and metazoan for the TAILs and IDOLs. Multiple sequence alignments (MSA) by ClustalW were used for constructing the Phylogenetic trees. Local alignment searches are from NCBI-BLAST. BLAST was activated with a 'gap costs' for Existence: 10 and for Extension: 1. The resetting of the BLAST parameters was needed for systematic identification of missing domains detection scheme. The phylogenetic trees were built using the iTol [56] . Table S3 ). The graphs show the distribution of averages proteins length (two distributions per each Pfam family: one for the metazoan proteins and one for the viral proteins. A statistical KS test was performed on the domains length. No significant difference between the metazoan domains and the counterpart viral domains is detected. The same results were observed when using other statistical tests (e.g., t-test, not shown). The average and median proteins length and the average and median domain length is shown, next to the results of the statistical significant tests. (PPT) Figure S3 Phylogenetic tree of the viral B1R kinase family. A BLAST search (http://blast.ncbi.nlm.nih.gov) for the 32 highest scored proteins that belong to the B1R kinase family is shown. The query protein used is protein kinase CMLV190 from Camelpox virus. All viruses that were identified belong to dsDNA Class I of different genera. The tree branches are color coded for viruses and mammals (including platypus). All the 21 viral sequences belong to dsDNA Class I from different genera. Representatives are of Orthopoxvirus (Variola, cowpox virus) Capripoxvirus (e.g., Lumpy skin disease virus), Leporipoxvirus (Rabbit fibroma virus) and Yatapoxvirus (e.g., Yaba monkey tumor virus) and more. (PPT) Figure S4 Linker lengths in Pfam families that contain viral and metazoan proteins. The cumulative fraction function for all analyzed Pfam families for TAIL and IDOL sequences. A zoomed section of this graph is shown in Figure 6 . Viral proteins are marked in red and metazoan proteins in blue. (PPTX)

Treatment of Neuroterrorism

Bioterrorism is defined as the intentional use of biological, chemical, nuclear, or radiological agents to cause disease, death, or environmental damage. Early recognition of a bioterrorist attack is of utmost importance to minimize casualties and initiate appropriate therapy. The range of agents that could potentially be used as weapons is wide, however, only a few of these agents have all the characteristics making them ideal for that purpose. Many of the chemical and biological weapons can cause neurological symptoms and damage the nervous system in varying degrees. Therefore, preparedness among neurologists is important. The main challenge is to be cognizant of the clinical syndromes and to be able to differentiate diseases caused by bioterrorism from naturally occurring disorders. This review provides an overview of the biological and chemical warfare agents, with a focus on neurological manifestation and an approach to treatment from a perspective of neurological critical care. ELECTRONIC SUPPLEMENTARY MATERIAL: The online version of this article (doi:10.1007/s13311-011-0097-2) contains supplementary material, which is available to authorized users.

Substances that can potentially be used as weapons of mass destruction or agents of terrorism may be chemical, biological, nuclear, and radiological [1] . Bioterrorism is defined as the intentional use of these substances to cause disease or death in humans and/or animals, and/or environmental damage [2] . In case of an attack, a large number of victims could be affected in a very short period of time, putting an enormous strain on the healthcare system [3] . Personnel will be faced with enormous logistical problems, and medications and other resources are likely to be insufficient [4] . Therefore, the United States (U.S.) Centers for Disease Control (CDC) urges healthcare professionals to be familiar with warfare agents, and in conjunction with governmental organizations have an implemented "Bioterrorism Preparedness and Response Program" to quickly detect and appropriately respond to a potential bioterrorist attack [5] . Early recognition is key in minimizing casualties, initiating appropriate therapy, and preserving resources. However, symptoms and signs caused by those warfare agents are often nonspecific and can easily be mistaken for common diseases. An important concept in differentiating a naturally occurring epidemic from a terrorist attack consists of recognizing an epidemiologic pattern [6] . Clues that suggest an attack include unusual age distribution or clustering of an illness [7] , a rapidly increasing incidence of an illness [8] , as well as an increased occurrence of an unusual illness or death in animals. The range of biological agents or chemical substances that potentially could be used as weapons of mass destruction is wide. The ideal agent can be produced and stored easily, in adequate amounts that are easy to disseminate, capable of producing a disease in great proportion to the exposed, and remains effective, despite environmental exposure and change of environmental conditions, and is challenging to detect [9] . Very few agents have all of these characteristics [10] . The CDC classifies potential bioterrorism agents into 3 categories: 1) A, 2) B, and 3) C. These categories are based on the agents' potential as weapons, such as their ability to be disseminated, transmitted, and to cause disease; the mortality rate; the expected impact on public health; and the potential for panic and social disruption [11, 12] . Category A agents, judged to have the greatest risk, include anthrax, plague, tularemia, smallpox, the hemorrhagic fever viruses, and botulinum toxin [11] . Most experts in the field believe that anthrax and smallpox would be the agents most likely to be used by terrorists [13] . Although these are the most easily fatal, terrorists could also reach their goals by simply causing illness on a large scale [10] . Category B agents are ones that would cause moderate morbidity and low mortality. Category C agents are pathogens in the emerging phase, [10] ( Table 1 ). The relative toxicity of selected agents for comparison is shown in Table 2 . Chemical and biological weapons can cause a wide range of nervous system damage and neurobehavioral effects. Therefore, preparedness among neurologists is as important as it is for emergency, infectious diseases, and critical care personnel [14] . The main challenge is to be cognizant of the clinical syndromes and to be able to distinguish diseases caused by bioterrorism from more commonly occurring natural disorders [15] . Nervous system complications in victims of warfare include penetration injuries to the brain and spine, contusions and concussions of the nervous tissue, meningitis and encephalitis, seizures, myelopathies, radiculopathies, peripheral neuropathies, post-traumatic encephalopathy, hypoxic brain injury, and behavioral changes [16] . Often, psychological symptoms would need differentiation from early manifestation of organic disease. In addition, vaccines against some categorized agents have neurological side effects (e.g., encephalitis after smallpox vaccination) [17] . In general, neurological disease tends to manifest somewhat later on in case of a biological attack, as compared to chemical weapons [15] . Prompt death, however, might occur following exposure to botulinum toxin, tetrodotoxin, saxitoxin, and nerve agents. Of the many agents that may be used, prominent neurological features occur with cyanide, cholinesterase inhibitors, botulinum toxin, anthrax [14] , and paralyzing toxins, as well as nerve agents (Table 3) . Anthrax is caused by Bacillus anthracis, a large, nonmotile, spore-forming, gram-positive rod. B. anthracis is common among domestic animals. It can be passed to humans by direct skin contact or inhalation of anthrax spores. Although the vegetative form survives poorly outside of a host [18] , the spore form can survive for decades [19] . It has many characteristics of an ideal biological weapon, its production is simple and cheap, and it can be stored for long periods of time. It is highly effective, with a morbidity rate of 65 to 80% if treatment is not promptly initiated. Weaponized anthrax can be produced as insoluble, liquid slurry, or dry powder. Although the most likely method of deployment is aerosolization of dry spores [20] , contamination of food and water supplies is conceivable [21] . The most serious terrorist threat posed by anthrax is infection by inhalation. For humans, the dose sufficient to kill half of the exposed persons ranges from 2500 to 55,000 inhaled spores [18, 22] . According to a U.S. government estimate, the outdoor release of 100 kg of B. anthracis in Washington, D.C. could produce between 130,000 and 3 million deaths [23] . Anthrax has been weaponized at various times in the past, most recently in October and November of 2002 in the U.S., which led to 18 confirmed and 4 suspected cases of disease [24] . Infection is acquired by ingestion, inhalation, or absorption of the spores through breaks in the skin and mucous membranes. Depending on the route of exposure, cutaneous, gastrointestinal (GI), or inhalation anthrax ensues. Most naturally occurring human infections are cutaneous from contact with infected animals or contaminated material. Naturally occurring inhalational anthrax is rare, particularly in the industrial world [25] ; therefore, the occurrence of anthrax should raise concerns of an intentional dissemination. Cutaneous transmission of the hands, arms, and face are the most common routes of clinical infection in humans. A pruritic papule evolves into an ulcer, followed by the development of a large painless black eschar. The eschar dries and desquamates after 1 to 2 weeks. Painful lymphadenopathy and sepsis can arise. With treatment, local cutaneous anthrax has a mortality rate of less than 1%; if the disease becomes systemic, mortality may be as high as 20% [26] . GI anthrax, while not common, occurs naturally as a result of ingesting poorly cooked, contaminated meat. Ulcers in the mouth or esophagus, or lesions lower in the intestinal tract may develop, and presenting symptoms include nausea, vomiting, diarrhea, abdominal pain, or an acute abdomen progressing to a sepsis syndrome with high mortality. Inhalational anthrax follows the deposition of sporebearing particles into alveolar spaces. From there, they are transported to the mediastinal lymph nodes. Subsequent germination within the lymph nodes leads to a massive release of bacteria and toxins into the bloodstream. The incubation period is usually less than 1 week, but it can be as much as 6 weeks. Initial symptoms of the clinically and fairly consistent 2-stage disease are nonspecific, with fever, chills, myalgia, cough, and sore throat [26] . Substernal chest pain, dyspnea, abdominal pain, nausea, and vomiting are common. With disease progression for 2 to 3 days, severe pneumonitis develops, and abruptly, sepsis, hypoxemia, cyanosis, and shock follow. Prominent shortness of breath reflects thoracic lymphadenitis and mediastinitis rather than bronchopneumonia. However, inhalation anthrax can sometimes present without the usual symptoms of chest pain and shortness of breath [27] . Weaponized anthrax presents with these inhalational findings. However, the epidemiology of weaponized anthrax is similar to that of a single point toxin exposure, with those exposed starting to become ill in relatively large numbers during a short period of time. In the Sverdlosk (now known as Ekaterinaburg) accidental release of 1979, most of the 68 known victims became ill within 2 weeks of exposure [28] . Death ensues approximately 24 to 36 h after the appearance of respiratory distress, but sometimes it occurs within hours [26] . Untreated, mortality reaches 95%. Among the confirmed inhalational cases from the attack in the fall of 2001, the case fatality rate was 45% [29] . For diagnosis, the organism may be detected by cultures and gram stain of blood or aspiration of skin lesions. Sputum Although the primary clinical presentation is a systemic or pulmonary illness, which is unlikely to be solely or initially neurological [15] , all 3 forms of anthrax can be complicated by meningitis, mostly in the second stage of the disease [30] . The risk of hemorrhagic meningitis in cases of inhalational anthrax is estimated to be as high as 50% [18] . There was 20% of the known patients who developed meningitis after the mail-borne inhalational anthrax attack [29] . The most common neurological manifestations are headache and confusion [29] . In the closely studied cases in the U. S. in 2001, neurological abnormalities were noted in 80% [29] . Meningitis presents with fever, headache, nausea, vomiting, and altered mental status. Clinical signs include meningeal signs, long-tract signs, hyperreflexia, seizures, myoclonus, fasciculations, rigidity, stupor, or coma. Untreated, mortality is high, but early diagnosis and prompt initiation of antibiotics can halt disease progression. Cerebrospinal fluid (CSF) shows neutrophilic pleocytosis often greater than 500 ml, elevated erythrocyte count, and elevated protein [14] , which are findings similar to the profile expected in herpes simplex virus (HSV) encephalitis or subarachnoid hemorrhage [30] . Gram-stain shows copious largegram positive rods with or without endospores. Blood cultures are positive in most patients with meningoencephalitis. Neuroimaging reveals diffuse cerebral edema, prominent leptomeningeal enhancement, focal intracerebral, subarachnoid, or intraventricular hemorrhage [31] . An electroencephalogram may show disorganized, low-amplitude slow waves of 1 to 7 Hz. At autopsy, the meninges show extensive fresh hemorrhage, sometimes described as a "cardinal's cap" [32] . Unless engineered, B. anthracis is susceptible to penicillin, amoxicillin, chloramphenicol, doxycycline, erythromycin, streptomycin, ciprofloxacin, and other quinolones. It is resistant to ceftriaxone and other 3 rd generation cephalosporins. Treatment for anthrax consists of a multi-drug regimen of ciprofloxacin, and at least one other agent of vancomycin, chloramphenicol, or penicillin [14] . For inhalational anthrax, the recommended regimen is ciprofloxacin or doxycycline, plus clindamycin and rifampin. Doxycycline and clindamycin, however, exhibit poor cerebrospinal fluid penetration and should be avoided in cases of anthrax meningoencephalitis. The addition of rifampin serves for the prevention or treatment of neurological manifestations [14] in cases treated with doxycycline or clindamycin. Treatment duration is long; a 60-day course is not unusual, given that spores can remain dormant for a long time. Corticosteriods are recommended in all patients who have pulmonary edema, respiratory failure, and meningitis [12] . In anthrax meningitis, steroids have been reported to improve survival [26] ; however, their use is controversial in adults, but it has improved outcome in children. Mortality rates of as much as 20% for the cutaneous form, 60 to 80% for the GI form, and 90 to 99% for the pulmonary form make prompt treatment essential [33] . With multi-drug antibiotic regimens and supportive care, survival rates have improved. If there is a delay in treatment initiation from 2 to 4.8 days, the mortality would be expected to double [34] . Vaccination is available for military personnel and civilian workers at risk for exposure [22] . The 2 types of vaccines for humans are both directed against the protective antigen of B. anthracis, and should protect against cutaneous and inhalational anthrax. It is given in 6 does of 0.5 ml for 18 months, followed by yearly boosters. Although the incidence of adverse reactions is low [18] , neurological side effects, such as optic neuritis have been reported [35] . If a bioterrorist attack is suspected, or after an exposure, prophylaxis with ciprofloxacin 500 mg twice a day or doxycyline 100 mg twice a day is recommended for the target population. Treatment duration should be 4 weeks, while the effects of simultaneous vaccination take effect [36] . Resistance to penicillin and tetracycline should be assumed until proved otherwise by susceptibility testing [20] . The differential diagnosis for anthrax includes mycoplasma pneumonia, Legionnaire's disease, psittacosis, tularemia, Q fever, viral pneumonia, histoplasmosis (fibrosing mediastinitis), and coccidioidomycosis. The spores can be inactivated in water of near boiling temperature (25 minutes at 95°C). Formaldehyde or 5 to 10% chlorine bleach can be used to destroy spores on contaminated surfaces [37] . Filtration removes spores if the pore size is less than 1micrometer μm. Given that there is no person-to-person transmission, standard infection control precautions are sufficient. In case of contact with spores, vigorous washing with soap and water is recommended, and the affected clothing should be placed in a plastic bag. The plague is caused by the gram-negative bacillus Yersinia pestis. It is a zoonotic infection of rodents that can be transmitted through flea bites, but also person-to-person. The plague is more difficult to use as a biological weapon than anthrax, because Y. pestis does not form spores, it is susceptible to drying, heat, and ultraviolet light, and does not survive well outside the host body. Therefore, so far there has not been an effective bioweapon using aerosolized bacteria [38] . Unlike anthrax, secondary cases may result from person-to-person transmission [10] , however this requires close contact with a patient during the final stage of the illness [39] . Experts believe that the danger of terrorists using this organism may be greatly exaggerated [40] . The plague can manifest in 3 different major forms: 1) bubonic, 2) pneumonic, and 3) septicemic. The bubonic plague begins as painful adenopathy 2 to 10 days after the infecting flea bite [41] , usually in the groin, axilla, or cervical region. A bubo is a 1-to 10-cm large, acutely swollen, erythematous, extremely painful, lymph node with surrounding edema and warmth [42] . Fever, chills, headache, and weakness occur with acute onset, and can transition to the septicemic form of the plague [42] in a quarter of patients. There are 80% of patients with the bubonic plague, which are bacteremic. There are 5 to 15% of bubonic plague victims who develop pneumonic plague, and hence become contagious. The overall mortality is estimated to be 60%, but can be less than 5% with prompt initiation of treatment. The pneumonic plague usually manifests after an incubation period of 2 to 3 days with fulminant pneumonia, malaise, high fever, cough, hemoptysis, and septicemia with ecchymoses, and extremity necrosis. The disease progresses rapidly, leading to dyspnea, stridor, cyanosis, and septic shock. Death is normally the result of respiratory failure and circulatory collapse [42] . The pneumonic plague is highly contagious via inhalational exposure or secondary hematogenous spread, and therefore the most likely form to be used in a bioterrorist attack. It is invariably fatal, unless treated within the first day of onset. Early diagnosis is important in initiating treatment within 24 h of symptom onset, which is crucial for survival. Aspiration of a bubo or sputum and gram stain analysis can provide a rapid bedside diagnosis. Definite diagnosis is made by culture; the cultures are often negative for 24 h, but turn positive at 48 h. The anti-Y. pestis titer rises fourfold or greater. Blood count shows a leukocytosis with left shift, and bilirubin and aminotransferase levels are elevated. The central nervous system (CNS) manifestations with meningeal involvement can complicate any of the forms and occur in approximately 6 to 7% of plague cases. Cerebrospinal fluid analysis reveals a neutrophilic pleocytosis [43] . The differential diagnosis for the pneumonic plague includes disease caused by other biowarfare agents, such as anthrax, tularemia, and melloidosis (glanders), and other pneumonias such as severe community-acquired pneumonia, hantavirus pulmonary syndrome, influenza, or leptospirosis. The septicemic plague has to be differentiated from meningococcemia, Rocky Mountain spotted fever, other gram-negative sepsis, and thrombotic thrombocytopenic purpura. Symptomatic patients should be isolated with strict respiratory isolation until treatment for at least 3 days [10] . The treatment of choice is streptomycin, alternatively doxycycline, gentamicin, ceftriaxone, chloramphenicol, or fluoroquinolones can be used. Treatment duration is for 10 days at a minimum. If exposed to aerosolized plague or to a patient with suspected pneumonic plague, prophylaxis with ciprofloxacin, doxycycline, tetracycline, or chloramphenicol should be given [42] . Francisella tularensis is a nonmotile, aerobic, gramnegative coccobacillus [44] . There are 4 subspecies. Usually it is associated with zoonoses in rural areas [8] . In North America, type A, which is believed to be the most virulent strain, is predominant [45] . F. tularensis is highly infectious: only 10 to 50 organisms are needed to cause human disease [8] if inhaled or injected. Oral ingestion requires approximately 108 organisms that lead to disease. Human-to-human transmission has not been reported. The most likely method of deployment, therefore, would be made via aerosol, although contamination of food and water sources seem possible [8] . In a World Health Organization report from 1969, it was reported that 50 kg of aerosolized F. tularensis in an area inhabited by 5 million people would result in 19,000 deaths and 250,000 persons with severe illness [46] . F. tularensis can survive for weeks in the environment and for years in temperatures of freezing and below [10] ; however, it is easily destroyed by heat (55°C for 10 minutes) or standard disinfectant solutions, such as 10% bleach [8] . The clinical manifestations depend on the route of infection. It can be transmitted through a bite from an infected arthropod or handling of infected animal carcass, ingestion of contaminated food or water, or inhalation of droplets [44] , and respectively patients can present with ulceroglandular, glandular, oculoglandular, oropharyngeal, typhoidal, or pneumonic tularemia [45] . The incubation period usually comprises 3 to 6 days. Ulceroglandular tularemia, which is the most common form and makes for 80% of patients, starts with the infected suppurative skin lesion, most commonly the hands, and localized lymphadenopathy. The original skin lesion erupts and ulcerates with raised edges. Glandular tularemia in confined to lymphadenopathy [47] . Oculoglandular tularemia ensues after inoculation of the organism through the conjunctiva, with painful conjunctivitis, and preauricular, submandibular, and cervical lymphadenopathy. Oropharyngeal tularemia occurs after consuming contaminated food, with painful exudative pharyngitis and tonsillitis [44, 45, 47] . Pneumonic tularemia is similar to an atypical pneumonia with abrupt onset of constitutional symptoms and a nonproductive cough [48] . Typhoidal tularemia is the systemic form that occurs in 30% of cases after any form of acquisition, but most commonly after inhalation of infectious aerosols. From the regional lymph nodes, the organisms spread to various organs, such as the liver, spleen, lungs, kidneys, intestines, CNS [44, 45] . It would be the most likely form to be encountered after use of francisella tularensis as a bioweapon. It is characterized by high fevers, headache, myalgias, prostration, vomiting, diarrhea; renal failure, rhabdomyolysis, pericarditis, meningitis, and erythema nodosum [47] . Approximately 80% of patients have pneumonia. Neurological manifestations with severe meningitis or encephalitis are rare and only occur with widespread dissemination and sepsis [48] . Case fatality rates of untreated naturally acquired typhoidal cases is approximately 35% compared with 1 to 3% for appropriately treated cases [10] . Diagnosis is usually made by serology. A high antibody titer can be detected by enzyme-linked immunosorbent assay (ELISA), but is not very sensitive in the first week [49] . Titers become positive during the second week of infection in 50 to 70% of cases, and reach their highest level after 4 to 8 weeks [50] . Definitive diagnosis can also be made by culture of oropharyngeal specimens or fasting gastric fluid; however, the organism rarely can be isolated from blood [48] . PCR from wound swabs is 78% sensitive and 96% specific [51] . Treatment regimens according to the Working Group on Civilian Biodefense [48] are streptomycin (1 g intramuscularly twice a day×10 days) or gentamicin (5 mg/kg intravenously or intramuscularly every day×10 days) for isolated cases, and ciprofloxacin (500 mg by mouth twice a day×10 days), or doxycycline (100 mg by mouth twice a day×10-14 days) in the setting of a mass casualty. Postexposure prophylaxis is with ciprofloxacin or doxycycline for 2 weeks. Given that person-to-person transmission is rare, standard precautions are sufficient. Q fever is caused by the intracellular coccobacillus Coxiella burnetii [52] after exposure to infected sheep, cattle, goats, or other livestock [8] . The bacterium's spore-like form is resistant to heat and desiccation, and it can persist for months [8] . This form can be distributed easily by wind [8] . It is highly infective; only 1 to 100 organisms are necessary to produce disease [52] . It cannot be transmitted human-to-human, but tissue may pose a risk [10] . Exposed surfaces can be decontaminated with 5% hydrogen peroxide or 70% ethyl alcohol for 30 minutes [52] . The incubation period lasts from days to several weeks. The presenting symptoms are nonspecific; most patients experience a febrile flu-like illness with or without cough, which resolves within 1 to 2 weeks [8] . Neurological manifestations occur in as much as one fourth of patients, and include severe retrobulbar headache, meningitis, and encephalitis [53] . Mortality is reported to be 2.4% [54] . Chronic morbidity is low as well [8] ; however, endocarditis, intravascular infection, hepatitis, or osteomyelitis may persist. Diagnosis can be made by ELISA. Treatment options are tetracycline, doxycycline, or macrolides; fluoroquinolone are to be considered in meningitis [55] . Treatment should be continued until the fever has subsided for 1 week [52] . Postexposure prophylaxis with a 5-day course of tetracycline or doxycycline may be effective if initiated within 8 to 12 days of exposure [52] . Brucellosis is caused by Brucella species, small, aerobic, slow-growing gram-negative coccobacilli. There are 4 of 6 species (B. abortus, B. melitensis, B. suis, and B. canis) that can cause human disease. Brucella species can survive for many weeks in water or soil. It could be spread as a dry aerosol or in bomblets [8] . Infection occurs most often after ingestion of unpasteurized dairy products or contact with infected meat or animals [56] . Most infections remain asymptomatic. Depending on the organism, symptoms begin as early as 2 weeks after exposure, but can occur as late as months after exposure. The organism tends to seed tissues with large numbers of macrophages, such as lung, spleen, liver, CNS, bone marrow, and synovium. The disease most often starts with a nonspecific prodrome, which is, however, absent in infection with B. melitensis. This is followed by the bacteremic stage, with intermittent fever, lasting for several weeks before subsiding, and then recurring in addition to other symptoms. This pattern of periodic febrile waves and remission can last for months or even years. Common manifestations in naturally acquired disease include joint pain, which is often incapacitating, and most commonly affects the sacroiliac joint, but also ankles, knees, and hips. Low back pain is seen in 60% of infected people and can be associated with vertebral osteomyelitis, intervertebral disc, or sacroiliac infection, or paravertebral abscess. Although pneumonia is not a common complication of brucellosis, 20% of patients develop cough and pleuritic chest pain. GI symptoms develop in 70% of adult cases. Hepatomegaly or splenomegaly is the result of granuloma formation and occurs in 45 to 63% of cases [57] . Endocarditis occurs in fewer than 2% of cases. Neurobrucellosis with direct invasion of the CNS complicates less than 5% of infected individuals [58] . It may manifest as meningitis or meningoencephalitis, demyelination, cranial neuropathies, myeloradiculitis, cerebral arteritis, or spinal peripheral entrapment neuropathy [59] . Diagnosis can be made by blood culture, bone marrow aspiration, or serology. In patients with neurological symptoms, CSF analysis reveals a lymphocytic pleocytosis and elevated protein. CSF cultures are positive in 13% of cases [60] . Although most patients will recover without treatment, antibiosis reduces the severity and duration of the disease. The most commonly used regimen consists of doxycycline plus rifampin for 6 weeks, but up to 3 to 4 months. Gentamicin or streptomycin is sometimes added in more severe infections [10] . Steroids may be beneficial in patients with encephalitis or meningitis. There is no human vaccine available for brucellosis. The mortality rate for untreated brucellosis is estimated to be 5%; death occurs in severe cases with meningitis or endocarditis. Glanders is caused by the nonmotile gram-negative bacillus Burkholderia mallei. Due to its ability to result in serious infection and the possibility of it being spread through aerosol, B. mallei may have potential as a bioweapon [10] . Infection from inoculation through skin break typically results in a tender nodule with local lymphangitis. If transmitted through mucosa of the eyes, nose, or oropharynx, mucopurulent discharge with ulcerating granulomas may occur. If inhaled and causing systemic invasion, septicemia develops after 1 to 2 weeks, and the disease commonly manifests as pneumonia [61] . The most common manifestations include fever, myalgias, headache, and pleuritic chest pain. Lymphadenopathy or splenomegaly can often be found. A generalized papular or pustular rash is frequent. The septicemic form frequently results in death within 7 to 10 days. Distinct neurological manifestation is not expected, but nonspecific symptoms, such as headaches are encountered as part of the common manifestation. The organism is difficult to identify. Cultures usually remain negative. Antibiotics used to treat human melioidosis include tetracyclines, trimethoprim, and sulfamethoxazole, amoxicillin clavulanate, and chloramphenicol. Strict isolation of infected patients is indicated due to the possibility of person-to-person transmission. Smallpox is caused by a DNA virus of the orthopox family. It can be transmitted by aerosols, droplets, direct contact with infected skin lesions, or even contaminated clothing or linens, and spreads easily from person-to-person [62] . Humans are the only reservoir for the virus [8] . Smallpox was declared eradicated by the WHO in 1980 [62] , and routine vaccination was stopped soon afterward. The virus is officially stored at 2 laboratories of the WHO, in the U.S. and in Russia [8] , although it is possible that clandestine samples are held elsewhere. As aerosolized smallpox is extremely virulent with a low infectious dose and the easy transmission from person-to-person even in asymptomatic stages, smallpox is 1 of the most feared agents that could be used in a biological attack [62, 63] . Smallpox infection occurs as major and minor form (variola maior, variola minor). The major form has 3 clinical phases: 1) the incubation period, 2) a prodromal illness, followed by 3) a fulminant infection [63] . The asymptomatic period lasts from 7 to 17 days (usually 12 to 14 days) after the initial exposure [62] . Asymptomatic viremia develops 3 to 4 days after infection. After multiplication of the virus in the spleen, bone marrow, and lymph nodes, a secondary viremia develops on approximately day 8 of infection [62] . During this prodromal phase, nonspecific symptoms, such as malaise, headache, backache, myalgias, fever, and vomiting develop. The overt smallpox syndrome occurs 2 to 3 days later, while the prodromal symptoms are subsiding. Infected leucocytes transport the virus to dermis and oropharyngeal mucosa, leading to the characteristic skin lesions [62] . Within 2 to 3 more days, a maculopapular rash appears; the greatest concentration of the lesions is in the face and distal extremities. The rash spreads from there in a centrifugal pattern [8] . Macules transform to papules to vesicles to pustules, each stage lasting 1 to 2 days. Vesicles and pustules are deep-seated, firm, round, well-circumscribed lesions; they are sharply raised and feel like small round objects embedded under the skin. Eventually, the lesions crust over and form scabs, leaving deep pitting scars that are unique to variola. Unlike varicella, all smallpox lesions are at the same stage of development. A more fulminant form, hemorrhagic smallpox or blackpox, occurs in approximately 3 to 10% of cases. The incubation period is shorter, and the characteristic rash presents as a dark, dusky erythema followed by petechiae and frank hemorrhage into the skin and GI tract. This form is almost uniformly fatal [64] ; death occurs 5 to 6 days after the onset of the rash [62] . This illness could be confused with meningococcemia or acute leukemia. During the phase of the rash, patients are most infectious as virus particles are released from the lesions or infected mucosa [62] . Patients stay contagious until all scabs separated [62] . Infectivity is low during the incubation period and the first 2 days of fever and increases during the febrile period. Carriers can even be asymptomatic, shedding infectious virions without ever manifesting the disease [8] . Complications of smallpox infections include panophthalmitis, keratitis, corneal ulcers, blindness, osteomyelitis, arthritis, orchitis, and encephalitis [65] . Delirium occurs in approximately 15% of patients [8] . Encephalitis is reported to occur in 1 of 500 cases of variola major, and 1 of 2000 of variola minor, usually developing during the stages of the rash. Psychosis and seizures may occur [66] . Mortality is reported as approximately 30% for variola major among unvaccinated persons, but this reflects historical data. Mortality in the minor form is less than 1% [66] . Previously vaccinated patients experience a milder disease, a shorter course, and a lower mortality rate. Diagnosis is usually clinical, but must be confirmed by laboratory testing. PCR, antibody detection, or virus isolation are possible. Specimens should be handled under biosafety level 4 conditions if smallpox is a consideration [10] . The most important aspect, once the disease is suspected, is prevention of further disease spread by strict isolation of patients and quarantine with respiratory isolation for 17 days of people with direct contact to patients. In patients with neurological complications, CSF usually shows a neutrophilic pleocytosis by day 2 to 4, which later turns into a lymphocytic pleocytosis. Treatment is mostly supportive. Cidofovir has shown antiviral activity in vitro, but is not approved for use in humans with smallpox [67] . Unlike many other vaccines, the smallpox vaccine can be effective in preventing or attenuating disease, even when administered within 4 days after exposure [62] . As vaccinia is a live virus, secondary transmission after vaccination is possible. The vaccine provides 90 to 97% protection for at least 3 years. Smallpox vaccination is not without risk. There may be cardiac adverse events, so the vaccine is not recommended for people with cardiac disease. Workers in the former Soviet Union developed a weaponized form of smallpox in which the onset of the disease is shortened, decreasing the likelihood that postexposure vaccination would be effective [68] . The most feared complications are CNS complications, such as encephalitis and encephalopathy [69] , which occur in 1 in 100,000 to 500,000 [67] . Postvaccinal encephalitis presents with headache, meningismus, fever, drowsiness, and vomiting; some cases are accompanied by spastic paralysis. A second form, postvaccinal encephalomyelitis, may present in 11 to 15 days after vaccination, similar to encephalitis with fever, mental status change, meningeal signs, seizures, and additional spinal cord dysfunction. Mortality of these complications is as high as 25% [62] , and 25% of survivors develop persistent deficits [67] . Viruses that cause hemorrhagic fevers and are category A agents in the CDC classification are the Ebola, Marburg, Lassa, Junin, Machupo, Guanarito, and Sabia viruses [70] . They are widely distributed in nature. Humans are highly susceptible [71] . Many are spread by airborne transmission, and although humans are not natural hosts for any of the viral hemorrhagic fevers, infected humans can spread the disease from person-to-person [12] . All of those cause fever, malaise, vomiting, and may evolve into diffuse hemorrhage and bleeding diathesis [10] , but they all have a unique set of clinical complications [70] . The incubation period varies from 4 to 21 days until the nonspecific prodrome develops. Within hours or days after initial presentation, the clinical condition rapidly deteriorates, which results from the affinity for the vascular system of the virus. Increased vascular permeability leads to flushing, petechial hemorrhages, mucus membrane hemorrhage, and shock, often with neurological, pulmonary, or hepatic involvement [64] . Signs of CNS involvement, such as delirium, seizures, or coma, usually indicate a poor prognosis. Patients who survive this disease may be left with hearing or vision loss, impaired motor coordination, transverse myelitis, uveitis, pericarditis, orchitis, parotitis, hepatitis, or pancreatitis. Laboratory evaluation shows thrombocytopenia, disseminated intravascular coagulation (DIC), elevated liver enzymes, and elevated creatinine. A diagnosis can be made by ELISA in specialized laboratories. Treatment is mainly supportive. Infection control includes contact precautions and careful handling of all bodily fluids. Ribavirin is effective against arenaviruses (Lassa and New World arenaviruses) and bunyaviruses (Rift Valley fever, Crimean-Congo hemorrhagic fever, and Hantavirus) [64] . Alphaviruses are categorized as category B agents by the CDC, as they are stable during storage and can be fairly easily produced in large amounts [8] . Diseases caused by alphaviruses are mainly neurological and include Venezuelan equine encephalomyelitis and Eastern and Western equine encephalomyelitis. This disease occurs naturally in North, Central, or South America, but human illness is rare, and most infections result in nonspecific symptoms of fever, headache, and myalgia. Less than 6% of infected adults or children will develop encephalitis, however the mortality rate of those can be as high as 50 to 75% for Eastern equine encephalitis [72] , which is the most severe of these infections, and survivors frequently have neurological sequelae [73] . Diagnosis is made by serological testing of CSF or serum. Treatment is supportive. There is no person-to-person spread. Venezuelan equine encephalitis virus is an alphavirus that is most commonly found in Central and South America. It is transmitted to humans by mosquitoes. In case of a bioterrorist attack, the distribution would be made through aerosols [17] . The virus usually leads to an initial severe febrile illness in nearly everyone exposed at 1 to 6 days after exposure. Naturally, only few patients (4% in children and less than 1% in adults) develop a severe encephalitis in a second phase a few days later [74] , but in case of an attack, increased numbers of encephalitis cases would be expected. Diagnosis is made by isolation of the virus in serum or throat culture. CSF shows a pleocytosis. Viremia is typically absent in patients with encephalitis. Preventative and postexposure treatments are limited. Vaccines that have been shown to have some protective efficacy [75] are available for laboratory personnel at high risk of exposure. Pegylated interferon-α (IFN-α) improves survival in mice [76] , but data for humans are not available. The overall mortality rate in a natural epidemic is estimated to be less than 1%, however, this increases to 20% if encephalitis develops [77] . Botulinum toxins are the most toxic substances known, and thus a potentially devastating weapon if efficiently dispersed [1] . The lethal dose of botulinum toxin for a 70 kg human is estimated to be 0.7 to 0.9 μg inhaled or 70 μg ingested [78] . Enough toxin is present in a single gram of crystallized botulinum toxin to kill more than 1 million people [14] . It is 15,000 times more lethal than the highly potent chemical agent VX and 100,000 times more lethal than sarin [8] . Botulinum toxin is produced by the obligatory anaerobic, gram-positive spore-forming soil bacterium Clostridium botulinum, and some strains of C. baratii and C. butyricum. There are 7 types of botulinum toxin (A-G), all of which use the same mechanism of action and can cause botulism. The toxin subtype is A in 50%, the remainder is usually B or E [79] . Types A, B, E, and F cause human disease, primarily affecting the nervous system [80] , and thus are of importance to neurologists. The toxin is readily absorbed by mucosal membranes, but it does not penetrate intact skin [14] . The bloodstream carries the toxin to the peripheral cholinergic synapses. It enters neurons by endocytosis at the nerve terminal and prevents synaptic vesicles from fusing with the nerve terminal, preventing their release of acetylcholine [81] . As few as 10 molecules of botulinum toxin can irreversibly stop acetylcholine release. The result is complete failure of neuromuscular junction transmission, followed by degeneration of the motor end plate and denervation of the muscle fiber. Most cases of naturally occurring botulism result from the ingestion of improperly prepared or inadequately homecanned food [82] . Although rarely, the disease is also associated with infected wounds or abscesses related to injection drug use. In infants, the toxin can be produced during growth of C. botulinum in the bowel. There is no natural inhalation botulism. The toxin is colorless and odorless, such that terrorists could contaminate food supplies [78] . Aerolization of preformed botulinum toxin is believed the most likely means of deployment of botulinum toxin in a warfare scenario [78] . Despite its high toxicity, the toxin is easily destroyed by heat; a temperature of 80°C for 30 minutes or 85°C for 5 minutes effectively degrades and inactivates the toxin [78] . Decontamination of exposed objects can be accomplished by washing them in a 0.5% sodium hypochlorite solution [83] . As there is no person-to-person toxin transmission, standard precautions are sufficient when caring for exposed individuals. Botulism has a characteristic presentation [10] . Unlike other threat toxins, botulinum toxin appears to cause the same disease independent from its route of exposure. The neurological syndrome is caused by presynaptic blockade of neuromuscular and autonomic cholinergic junctions [1] . The time of onset of symptoms varies with route of intoxication, and it is also dose-dependent [8] . Incubation time following ingestion is 12 to 36 h, with a range from 2 h to 8 days [79] . Symptoms after inhalation usually start 18 to 72 h after exposure. The rapidity and severity of paralysis depends on the amount of toxin absorbed. The clinical hallmark of botulism is an acute, afebrile, descending, symmetric, flaccid paralysis that always begins in the bulbar musculature [78] . Cranial nerve palsies invariably occur, making bulbar symptoms, such as ptosis, diplopia, dysphonia, and dysarthria, some of the earliest and most indicative symptoms of contamination [79] [14] . The earliest clinical signs are usually blurred vision from dilated pupils, ptosis, dry mouth, dysarthria, and dysphagia, as well as generalized weakness, fatigue, and dizziness. By the third day after exposure, patients will pool mucous in the throat, experience difficulty swallowing solid food, and have a sense of catching a cold, but without fever. Bilateral facial palsy is common. The cranial nerve palsies are followed by a symmetric, descending paralysis of skeletal muscles, which can quickly lead to respiratory failure. Severe weakness tends to occur by day 4 after exposure. Pharyngeal and upper airway paralysis may result in obstruction, and diaphragmatic and accessory muscle paralysis may render ventilation inadequate [78] . Death is usually a consequence of respiratory muscle failure or upper airway obstruction. Ascending weakness has not been reported. True sensory changes are not encountered, but hyperventilation may produce paraesthesias. Patients remain fully conscious, as the toxin does not penetrate the blood brain barrier; however, mental numbness may occur, and patients may appear lethargic because of diffuse muscle weakness and difficulty communicating due to bulbar weakness [78] . Urinary retention or GI ileus may occur with abdominal cramping. Postural hypotension may be present. Deep tendon reflexes are intact in the beginning, but decline during a period of days. There are no dermatologic abnormalities. The classic triad of botulism, according to the Working Group on Civilian includes [1] symmetric, descending flaccid paralysis with prominent bulbar palsies in [2] an afebrile patient with [3] a clear sensorium [78] . The diagnosis of botulism is primarily clinical. Descending paralysis with prominent cranial nerve involvement and autonomic dysfunction (especially the gastrointestinal (GI)) should raise suspicion [15] . CSF and routine blood studies are typically normal, as are imaging studies of the brain, and thus they have limited value in the acute setting [83] . Definitive diagnosis requires detection of botulinum toxin in serum or stool, gastric aspirate, and if possible the suspected source [78] . For serum confirmation, testing must be done on ≥30 ml of blood in adults before therapy with antitoxin. However, toxin in serum or stool is identified in less than half of clinically diagnosed cases [79] . A mouse bioassay is the standard laboratory diagnostic method, in which the toxin type is identified by protecting mice with specific antitoxins against individual strains. The test takes days to be arranged and performed; therapy and notification of public health authorities must be based on clinical suspicion [78] . An antibody response is not mounted in most patients, because the amount of toxin required to produce a clinical syndrome is not large enough to generate an immunological response [8] . On electrophysiological testing, motor conduction velocities and sensory nerve conduction remain normal. Compound muscle action potentials from affected muscles are diminished [84] . High frequency repetitive nerve stimulation produces an incremental muscle response similar to the Eaton-Lambert syndrome [85] . Autonomic function studies show an absent sympathetic skin response and significantly decreased heart rate variation [86] . Differential Diagnosis. Botulism may be confused with Guillain-Barré syndrome (especially the Miller Fisher variant), myasthenia gravis, or a pontine stroke. Furthermore, the differential diagnosis includes drug intoxication, poliomyelitis, tick paralysis, diphtheria, and paralytic shellfish poisoning. Botulism and atropine poisoning can both cause dilated pupils, dry mouth, constipation, urine retention, and prompt vomiting after food ingestion. The only specific treatment for botulism is passive immunization with an equine antitoxin. A trivalent antitoxin, which is active against the 3 most common types of food borne botulism (A, B, and E) is available from the CDC [8] . A pentavalent toxoid vaccine for types A, B, C, D, and E is only available to military personnel [8] . The U.S. Army possesses limited quantities of a heptavalent antitoxin, which might be available in a terrorist attack [87] . Although the antitoxin does not reverse existing symptoms, the deficits may stabilize and stop progressing [14] . Retrospective studies showed that early administration (within 24 h of symptom onset) reduced mortality and duration of hospital stay [88] . Animal studies suggest, if administered before clinical effects appear, the antitoxin might prevent symptoms from occurring [8] . The antitoxin is not generally recommended if a patient's exposure is greater than 72 h before administration [78] . The antitoxin is provided in a 10 cc vial that provides 5500 to 8500 international units of each type of specific antitoxin. It has to be diluted 1:10 in isotonic sodium chloride solution and must be slowly infused intravenously. Because it is of equine origin, hypersensitivity reactions are possible, and antitoxin administration should be preceded by a small challenge dose. Diphenhydramine and epinephrine should be available during administration of the antitoxin in case of a severe hypersensitivity reaction. Patients who respond to the test dose with a substantial wheal and flare can be desensitized for more than 3 to 4 h [78] . Antibiotics are not useful in the setting of inhalation or toxin ingestion, as it is not the bacterium itself, but the preformed toxin that is causing the illness [83] . Antibiotics may be useful for wound botulism, and for GI colonization with C. botulinum. The mainstay of therapy is supportive. Severe morbidity and death due to botulism is mostly attributable to aspiration or to respiratory failure. Close monitoring of cough and gag reflexes, assessment of oropharyngeal secretions, respiratory mechanics, and oxygenation is necessary. Mechanical ventilation should be strongly considered if the vital capacity falls below 15 ml/kg or negative inspiratory force measures less than 20 cm of water. Placement of a nasogastric tube to prevent aspiration and to permit nutrition in the setting of bulbar palsy often becomes necessary. When treating secondary infections, aminoglycosides and clindamycin should be avoided as they may exacerbate the existing neuromuscular blockade [78, 89] . Prognosis. Damage to the synapse and thus the neuromuscular blockade is permanent. Recovery only occurs with the sprouting of a new axon, which reinnervates the paralyzed muscle fibers [15] . In adults, this process may require many weeks or months or as much as a year or longer [14] . After many months, the original neuromuscular junction may regain activity. If respiratory paralysis has resulted, the patient usually remains ventilator-dependent during the recovery period, usually for 2 to 8 weeks [82] . In the case of a bioterrorist attack, supplying large numbers of patients with intensive care and mechanical ventilation would present tremendous logistical problems [10] . The fatality rate has been reported to be 25% for index patients and 4% for subsequently identified patients. Mainly, mortality is attributable to delayed recognition of the disease, or to the complications of prolonged intensive care [15] . Given that patients with botulism are not infectious to others, standard universal procedures but no barrier nursing are required [15] . Anatoxin A is a bicyclic amine produced by Anabaena flosaquae, a filamentous, freshwater bacterium found in pond scum worldwide [37] . A. flosaquae exhibits 2 mechanisms of action as an acetylcholine agonist by: 1) binding to postsynaptic acetylcholine receptors and 2) stimulating muscle contraction. As the binding to the receptor is permanent, continuous contraction of the affected muscle ensues [90] . Secondly, anatoxin A inhibits acetylcholinesterase, increasing the amount of acetylcholine in the synaptic cleft. Symptom onset occurs within a few minutes, and the combined effect of the 2 mechanisms of action of the toxins results in a flaccid paralysis [90] . Initially, symptoms may mimic organophaosphate poisoning, with miosis, excess oral and lacrimal secretions, and muscle fasciculations, [90] . Death results from respiratory arrest [37] . Supportive care is the mainstay of treatment. 2-pyridine aldoxime methyl (2-PAM) and physostigmine have shown some effect when used as pretreatment in animals [90] . During a terrorist attack, the toxin could conceivably be distributed by contamination of water supplies. Trichothecene mycotoxins are produced by the Alternaria, Fusarium, Aspergillus, Claviceps, Penicillium, and Stachybotrys species of fungi [37] . The best known toxin is T-2. Due to the simplicity obtaining the toxins, their resistance to autoclaving and ultraviolet light, and their rapid lethal effect, they have potential for use as biological weapons [91] . Inhalation, ingestion, or absorption through skin and mucous membranes leads to infection [91] . The toxins act by inhibiting protein synthesis and disrupting mitochondrial electron transport [37] . The main symptoms depend on the route of infection, and are cutaneous with blistering and skin necrosis, or respiratory with cough, dyspnea, and epistaxis. Neurological symptoms can include lethargy and incoordination. No rapid test is available for diagnosis; however, antigens and toxin metabolites can be detected in blood and urine within 1 month after exposure [91] . Treatment includes careful decontamination by washing with soap water, and 1% sodium hypochlorite solution with sodium hydroxide [91] , and is otherwise supportive. Ricin is a protein cytotoxin derived from the bean of the castor plant. Ricin acts by inhibition of DNA replication and protein synthesis, leading to cell death within 8 to 12 h [91] , and producing symptoms usually after 12 h [37] . Distribution of the toxin would most likely occur as an aerosol or droplet [91] . Clinical symptoms of the toxin depend on the route of exposure. Nonspecific symptoms include fever, nausea, arthralgias, and profuse sweating. After inhalation, chest tightness, cough, and dyspnea are prominent, and necrosis of the respiratory epithelium leads to tracheitis, bronchitis, bronchiolitis, and interstitial pneumonia [1] . When ingested, ricin causes nausea, vomiting, and diarrhea. If exposed to a sublethal dose, symptoms improve within several hours. Lethal doses produce necrosis of the respiratory tract and alveolar filling, or GI hemorrhage and hepatic, splenic, and renal necrosis [92] . Death from ricin toxin is dose-dependent, occurring 36 to 72 h after inhalation [1] . Death can be a consequence of pulmonary edema, acute respiratory distress syndrome (ARDS), disseminated intravascular coagulation, microcirculatory failure, or GI hemorrhage [93] . Injection of the toxin produces the most severe symptoms, and the CNS is affected early with convulsions [37] . Overall, the toxicity of ricin is much lower compared to botulinum toxin or Staphylococcus Enterotoxin B (SEB) [93] . The toxin can be inactivated by heat; 80°C for 10 minutes or 50°C for approximately 1 h is sufficient for neutralization of the toxin [37] . There is no specific treatment. This toxin is produced by the ubiquitous anaerobic, grampositive, spore-forming bacillus Clostridium perfringens. It can be found in the stool of every vertebrate. After accidental exposure, epsilon toxin causes increased vascular permeability leading to edema in various organs, and can result in a rapidly fatal acute toxemia. Inhalation can result in high permeability pulmonary edema, followed by circulatory spread with resultant renal, cardiac, and CNS damage. After ingestion, GI symptoms, such as watery diarrhea, nausea, and abdominal cramps will develop. Fever is rare. Spontaneous resolution typically occurs within a day. Fatality is rare, however, if delivered in high doses, epsilon toxin theoretically could rapidly debilitate civilian or military populations in large numbers [1] . There are at least 11 different enterotoxin serotypes, produced by various biotypes of Staphylococcus aureus. All subtypes are structurally similar and produce the same clinical syndrome [94] . Enterotoxin B is a potent T-cell activator, and the clinical symptoms are largely mediated by the immune system rather than direct toxic effects. The toxin is heat-stable and relatively stable in aerosols. It is the second most common cause of food poisoning, and when inhaled, even low doses can produce symptoms. Although the fatality rate is only approximately 5%, a high percentage of those exposed could become seriously ill within a few hours [95] . In naturally occurring disease, approximately 15% become ill enough to require hospitalization. Contamination of food or water supplies with enterotoxin could debilitate a population or army within hours [1] . Symptom onset is usually within 1 to 4 h, but can occur up to 12 h after exposure. Ingestions leads to nausea, vomiting, abdominal cramping, and diarrhea [10] . Less commonly, high fever, headache, myalgia, prostration, and dry cough develop. Symptoms resolve after a day [95] , but patients may be incapacitated for as much as 2 weeks. In severe cases, pulmonary edema or respiratory distress syndrome may develop. Death also may occur from dehydration [10] . Diagnosis can be made with a toxin assay. Treatment mainly consists of fluid and electrolyte replacement. There are 2 naturally occurring seafood neurotoxins: 1) tetrodotoxin produced by puffer fish, and 2) saxitoxin produced by microalgae in bivalve shellfish [37] . The toxins bind to voltage-gated sodium channels, inhibiting membrane depolarization and the conduction of action potentials [96, 97] . Both cause a severe paralysis of rapid onset. Numbness and tingling are often prominent, starting periorally before spreading to the limbs; GI distress, anxiety, headache, and mild peripheral weakness may appear within minutes to a few hours after ingestion. Successively, an ascending paralysis develops. Bulbar symptoms, hypersalivation, and sweating are commonly encountered. Hypotension (tetrodotoxin [97] ) or hypertension (saxitoxin [96] ), convulsions, and cardiac arrhythmias can occur. Death ensues secondary to respiratory failure within 24 h [37] . The victims may remain fully conscious. There is no specific treatment or antidote. Gastric lavage with activated charcoal and administration of anticholinergic agents has been suggested [98] . Intoxication can be survived with supportive treatment, as clearance of the toxin is fast. Recovery of survivors takes as much as 2 weeks [15] . The toxins can be deployed by contaminated food or water. They are not affected by temperature extremes and survive boiling [37] . Inactivation can be accomplished by chlorine under acidic and alkalinic conditions [37] . The toxins are highly potent, a thousand times more toxic than the chemical warfare agent sarin [99] . Inhalation is believed to produce the most severe effects [37] . Nerve agents are substances that cause their effects by inhibition of acetylcholinesterase and accumulation of acetylcholine. Medically used substances that cause these effects include carbamates (physostigmine, neostigmine, and pyridostigmine). In agriculture, insecticides (sevin) and organophosphates (malathion, diazinon) are used. The militarized nerve agents were originally synthesized as insecticides, before being used in World War II, and subsequently by Iraq against Iranian troops and Kurdish civilians, and by terrorists in Japan in 1994 in Matsumoto, and 1995 in Tokyo. They are the most toxic of the known chemical warfare agents. They are named Tabun, or "German agent A" (GA), "Sarin" (GB), Soman (GD), Cyclosarin (GF), and "Venemous" (VX) [1] . Their toxicity increases from GA to VX. They cause morbidity and mortality at extremely low doses [100] , persist in the environment for long periods of time, and can be released from contaminated clothing, skin, and secretions. The G type gases are clear colorless liquids, when fresh. VX is amber-colored and oily. Distribution occurs in gas form, with inhalation and absorption through the skin as the most common forms of intoxication [1] . They have no taste, and most are odorless; tabun has a slightly fruity odor, and soman's odor resembles camphor. The volatility is greatest for GB, followed by GD, GA, GF, and VX. Subsequent to binding to cholinesterase, sarin, soman, and cyclosarin lose fluorine; tabun, VX, and Russian VX lose cyanide and the thiol groups. The principal effect of nerve agents is exerted by inhibition of the enzyme acetylcholinesterase (AChE), which results in cholinergic overt stimulation with both muscarinic and nicotinic effects [101] . Pathophysiologically, their effects are the opposite of botulinum toxin; nerve agents result in increased acetylcholine in the synaptic cleft, while botulinum toxin results in decreased acetylcholine. The clinical manifestations of nerve agent intoxication are those of cholinergic excess. Muscarinic effects mainly manifest with symptoms from affected smooth muscles (Table 4 ) of airways, GI tract and eyes, glands, and the heart. Nicotinic effects concern skeletal muscles and pre-ganglionic nerves [1] . The mnemonic Salivation, Lacrimation, Urination, Defecation, GI hypermotility, Emesis (SLUDGE) summarizes the commonly experienced early symptoms of salivation, lacrimation, urination, defecation, GI hypermotility, and emesis [14] . The initial effects of nerve gas exposure depend on the dose and route of exposure. With exposure to vapor in small amounts, smooth muscles and glands of eyes, ear-nosethroat (ENT), and GI tracts and airways are mostly affected with miosis, rhinorrhea, salivation, and shortness of breath. The onset of those effects is within seconds to minutes. There is no worsening after the removal from the exposure, and no late-onset effects. After a large exposure to vapor, all symptoms of a small exposure are more prominent, and the CNS is affected. CNS symptoms range from irritability to convulsions and coma [102] . Nicotinic symptoms include weakness of skeletal muscles, fasciculations (localized in areas where droplets penetrated skin, generalized with respiratory or large transdermal exposures [102] ), and paralysis. Muscarinic symptoms include profuse exocrine secretions (tearing, rhinorrhea, salivation, bronchorrhea, and sweating), in addition to ophthalmic symptoms, such as miosis, dim vision, headache, and eye pain. Large doses may lead to seizures and coma. Cardiovascular effects initially are due to nicotinic stimulation, leading to tachycardia and hypertension [103] , but hypotension and cardiac conduction abnormalities are seen as well. Pulmonary symptoms include chest tightness, labored breathing, wheezes, and copious secretions. Acute respiratory failure is a combined effect from bronchoconstriction, marked increase in airway secretions, and respiratory muscle weakness. With dermal exposure, there is a delay of symptom onset for as much as several hours, and symptoms may persist even after decontamination due to the rapid absorption. The most sensitive indicator is miosis. Miosis is almost always present after vapor exposure and after large liquid exposure, and possibly after exposure to medium amounts of liquid nerve agents. High-dose exposure can produce rapidly (seconds to minutes) fatal systemic effects. If patients survive a large exposure because death from hypoxia is averted by atropine and 2-PAM, the CNS cholinergic effects become overt in form of convulsions. Seizures may evolve into status epilepticus, which can be prevented by giving large quantities of atropine early on Activity of plasma butyrylcholinesterase is more sensitive for most insecticides. Neurophysiological studies may assist in the diagnostic process. In acute organophosphate poisoning, nerve conduction velocities and distal latencies are normal, even in severely paralyzed patients [104] . The earliest and most sensitive indicator of the AChE inhibition is a small amplitude of compound muscle action potential after single supramaximal stimulation with often repetitive activity [104, 105] . On repetitive nerve stimulation, there is usually no decrement when stimulating at 3 Hz, and only occasional decrement at 10 Hz. At 30 or 50 Hz, there may be a decrement-increment response [104] in less severe stages of poisoning [105] . One of the most important principles in management of nerve agent exposure is self-protection with protective gear. Initial treatment of the victim consists of physical removal of clothing or other exposed objects, and decontamination and forceful wash with soap and water or 0.5% sodium hypochlorite [5] . Early skin decontamination, within 1 to 2 minutes, is best. There is little benefit after 30 minutes. The principle of antidotes is to reverse the effects of excess acetylcholine by inhibiting cholinergic effects and by reactivating the enzyme. Atropine may help to reverse bronchial constriction, which is given a starting dose of 2 to 6 mg followed by 2 mg every 5 to 10 minutes until the secretions halt and ventilation is improved. High cumulative doses (10-20 mg) in the first hours are not uncommon. Monitoring for atropine toxicity (delirium, hyperthermia, increased fasciculations) is necessary. Atropine may also cause arrhythmias and may even result in ventricular fibrillation if given intravenously in the presence of hypoxia. Electrocardiographic changes (ST depression and T-wave flattening) and cardiac arrhythmias reflect atropine toxicity and may be treated with propranolol. The combination of atropine with benactyzine is believed to be more effective, presumably by increasing central anticholinergic activity [5] . Oximes work by reactivating AChE. They bind to the organophosphate-inactivated AChE and displace and hydrolyze the organophosphate. They must be administered rapidly to be effective, due to a process called "aging," which refers to the organophosphoryl moiety and the amino acids of the active site becoming covalent and changing their structure. Once this has happened, the enzyme cannot be reactivated. The aging times depend on the nerve agent: GD has the fastest aging time, with a half-time of 2 minutes [106] . GB ages in 3 to 4 h, others take longer; VX ages very little. The oximes affect nicotinic sites; there is no clinical effect at muscarinic sites, and available oximes do not cross the blood brain barrier. Pralidoxime is the compound most frequently used in a dosing of 1 to 2 mg in 100 cc normal saline for 15 to 30 minutes, followed by a second dose after an hour if paralysis persists [102] . In critically ill patients, a pralidoxime infusion at 7.5 mg/kg/h is safe [107] . Very rapid administration of pralidoxime, on the other hand, can worsen motor weakness. Oximes are mostly given in conjunction with atropine and benzodiazepines. The dosing for 2-PAM chloride is simplified by combipen, which contains 600 mg. Infusion of intravenous doses of 25 mg/kg for approximately 25 minutes produces marked hypertension, which is rapidly but transiently reversed by phentolamine (5 mg). Apart from oximes, exogenous butyrylcholinesterase (Protexia TM) is available. For seizures, which may evolve into status epilepticus after pyridostigmine treatment leads to survival of an exposure, high quantities of benzodiazepines (usually diazepam in the military setting) may be required. The "convulsive antidote nerve agent autoinjector" (CANA) contains 10 mg diazepam. Pretreatment with physostigmine could help prevent the CNS consequences, but could also cause its own CNS toxicity. In animal studies of soman intoxication, ketamine in combination with atropine and benzodiazepines proved effective in stopping seizure, reducing brain damage, and increasing survival [108] The weakness usually resolves within 5 to 18 days. Ventilatory support in survivors is often required for several days or weeks. Table 4 provides an overview of the recommended therapy for casualties of nerve agents. Apart from the acute presentation, neurological sequelae of organophosphate poisoning may arise. A relapse of the weakness can occur 1 to 4 days after a seemingly welltreated and resolved course. This so-called intermediate syndrome has an incidence of 8% and presents with respiratory paralysis, cranial motor nerve palsies, and proximal limb and neck flexor muscles weakness [109] . Therapy is supportive, but patients may require (re)-intubation. Recurrent weakness typically resolves within 5 to 18 days [109] . Furthermore, an organophosphate-induced delayed polyneuropathy (OPIDP) may result from a distal dying back axonopathy [110] , believed to be caused by phosphorylation of the enzyme neuropathy target esterase (NTE) [111] . OPIDP appears 1 to 3 weeks after exposure with cramping pain in the legs, paresthesias, and motor weakness. This is rare after nerve agent exposure, but more common with insecticide overdose. Apart from neuropathy, pyramidal signs and symptoms can develop [112] . If exposure is low grade, but persistent, pervasive effects of nerve agent exposure on human emotion, learning, and memory may be ensue [113] . Gases, vapors, and other particles with a diameter of less than 2 mm can injure the entire airway [1] . Agents with highly water solubility (e.g., ammonia, sulfur dioxide) affect the upper airways with immediate burning sensation [114] , while low solubility agents (e.g., phosgene and nitrogen oxides), produce less immediate injury to mucous membranes and upper airways, and thus provide fewer warning signs of the exposure [114] . Massive exposure may lead to death from acute respiratory failure by destroying the alveoli and adjacent capillary endothelial cells. Delayed onset (up to 24 h) of acute lung injury is more common [114] . Vesicant agents are oily, clear to yellow-brown liquid alkylating agents. They lead to cell damage by alkylation of DNA [115] . Symptom onset ranges from 1 to 12 h after exposure in a dose-dependent fashion [116] , but it can be delayed. There is 20% absorbed from the skin [115] , and symptoms range from erythema and edema to necrosis and vesicles [116] . Groin and axillla are vulnerable due to their moisture and warmth [116] . Apart from the skin, the eyes and respiratory tract are affected [117] . Additional clinical effects include GI upset. Bone marrow suppression after high-dose exposure can be seen [117] . Long-term clinical consequences include blindness, chronic bronchitis, and cancers of the respiratory tract [118] . There is no known antidote. Fatality rates are low with 2 to 4% [116] . Sodium thiosulfate may prevent death by acting as a mustard scavenger if given within minutes of exposure. Hydrogen cyanide and cyanogen chloride are widely available, colorless, and come in gas or liquid form with high volatility. Hydrogen cyanide has an odor of bitter almonds; however, many people are not able to detect this distinctive odor [119] . Cyanogen chloride has a pungent, biting odor. They are absorbed through skin and respiratory mucosa. Mechanism of action is by interruption of the citric acid cycle and halting oxidative phosphorylation, inhibiting aerobic energy production and leading to rapid cell death [1] . Severity and types of symptoms depend on the level of exposure. Duration of exposure and ambient concentration of the substance influence whether a symptomatic threshold is reached. Therefore, the use of these agents in a terrorist attack is limited to a closed environment (e.g., an office space or a subway system) [14] . In confined spaces, these agents are highly lethal [1] . Mild exposure will cause headache, dizziness, drowsiness, mucosal irritation, and GI upset. Progression to coma can occur for several hours. Severe exposure leads to impaired consciousness and coma, arrhythmias, hypotension, cardiovascular collapse, respiratory irritation, and death [120] . The death can occur within minutes of inhalation [1] . Because hydrogen cyanide is excreted by the lungs, a patient's breath may have the characteristic of a bitter almond odor. The pupillary light reflex may be delayed [120] . Focal neurological signs are usually not prominent. Overall fatality rates are estimated at 11 to 34% [120] . If survived, sequelae are rare, but anoxic encephalopathy can occur [1] . Diagnosis should be suspected in an acyanotic patient with severe hypoxia. As differential diagnosis, carbon monoxide poisoning, exposure to organic solvents, drug intoxication, hypoglycemia, electrolyte disturbances, and postictal state should be considered. Cyanogen chloride induces mucosal irritation and excessive respiratory secretions, which are reminiscent of organophosphate poisoning. Laboratory findings are lactic acidosis and a decrease in the arterial-venous difference in partial pressure of oxygen [120] . Plasma thiocyanate levels can be measured, but not acutely. Treatment in mild cases can be limited to decontamination and observation and oxygen supplementation. In severe cases, treatment includes several antidotes, which are available in a prepackaged cyanide antidote kit [121] . Sodium thiosulfate promotes the formation of thiocyanate by the enzyme rhodanese and leads to excretion of thiocyanate in urine. Sodium nitrate and amyl nitrate lead to formation of methemoglobin, which has an affinity for cyanide, and thus helps to reduce its active presence. The desired methemoglobin level is between 20 and 30% of total hemoglobin [122] . Sodium nitrate is given intravenously, and amyl nitrate vapor can be administered by inhalation through saturated gauze or by emptying an ampule in a respirator reservoir. Delayed toxicity may affect the basal ganglia and present with Parkinsonian features. Dysarthria, eye movement abnormalities, dystonia, and ataxia have also been described [123] . Magnetic resonance imaging may show cavitation of the putamen and globus pallidus. Cortical, cerebellar, and diencephalic changes have also been reported. Disclaimer This article, including its tables, is intended to serve as a review of possible agents or biochemical warfare from a perspective of neurocritical care. It is in no way complete, nor is it intended to be complete. The appropriate agencies need to be consulted in case of a suspected attack or casualty.

New Cardiovascular and Pulmonary Therapeutic Strategies Based on the Angiotensin-Converting Enzyme 2/Angiotensin-(1–7)/Mas Receptor Axis

Angiotensin (Ang)-(1–7) is now recognized as a biologically active component of the renin-angiotensin system (RAS). The discovery of the angiotensin-converting enzyme homologue ACE2 revealed important metabolic pathways involved in the Ang-(1–7) synthesis. This enzyme can form Ang-(1–7) from Ang II or less efficiently through hydrolysis of Ang I to Ang-(1–9) with subsequent Ang-(1–7) formation. Additionally, it is well established that the G protein-coupled receptor Mas is a functional ligand site for Ang-(1–7). The axis formed by ACE2/Ang-(1–7)/Mas represents an endogenous counter regulatory pathway within the RAS whose actions are opposite to the vasoconstrictor/proliferative arm of the RAS constituted by ACE/Ang II/AT(1) receptor. In this review we will discuss recent findings concerning the biological role of the ACE2/Ang-(1–7)/Mas arm in the cardiovascular and pulmonary system. Also, we will highlight the initiatives to develop potential therapeutic strategies based on this axis.

The renin-angiotensin system (RAS) plays a key role in several target organs, such as heart, blood vessels, and lungs, exerting a powerful control in the maintenance of the homeostasis [1] [2] [3] [4] . This system is activated by the conversion of the angiotensinogen to the inactive peptide angiotensin (Ang) I through the renin action [5] . Subsequently, Ang I is cleaved by the angiotensin-converting enzyme (ACE) generating Ang-(1-7) by ACE2 is important to regulate the RAS activity since Ang-(1-7) induces opposite effects to those elicited by Ang II [16] [17] [18] [19] [20] [21] [22] [23] [24] . Additionally, ACE2 can form Ang-(1-7) less efficiently through hydrolysis of Ang I to Ang-(1-9) with subsequent Ang-(1-7) formation [24] . The relevance of the RAS is highlighted by the success obtained in therapeutic strategies based on the pharmacological inhibition of this system in cardiovascular and respiratory diseases [27] [28] [29] [30] [31] [32] . Blockade of the RAS with ACE inhibitors (ACEi) or AT 1 receptor antagonists (ARBs) improves the outcomes of patients with hypertension, acute myocardial infarction, and chronic systolic heart failure [33] [34] [35] . Furthermore, based on the involvement of the ACE/Ang II/AT 1 axis in respiratory diseases and the crucial role of the lungs in the RAS metabolism, several studies have reported the contribution of the RAS in lung pathophysiology [28, 30, 31, [36] [37] [38] [39] [40] . Importantly, it has been shown that administration of ACEi and ARBs causes substantial increases in plasma Ang-(1-7) levels, leading to the assumption that part of their clinical effects might be mediated by this heptapeptide [41] [42] [43] . Indeed, some effects of ACEi and ARBs can be blocked or attenuated by A-779, a Mas antagonist, confirming the role of Ang- (1) (2) (3) (4) (5) (6) (7) in the actions of these compounds [44] . The beneficial effects of Ang-(1-7), as well as its likely participation in the effects of the ACEi and ARBs, represent evidences for the potential of the ACE2/Ang-(1-7)/Mas axis as a therapeutic target. In this review, we will focus on the recent findings related to the pathophysiology actions of the ACE2/Ang-(1-7)/Mas axis in the cardiovascular and respiratory system. Also, we will discuss the promising initiatives to develop new therapeutic strategies based on this axis to treat pathological conditions. The heart is one of the most important targets for the actions of the ACE2/Ang-(1-7)/Mas axis. In the heart, ACE2 International Journal of Hypertension 3 is expressed in the endothelium [45] , myofibroblasts [46] , cardiomyocytes, and fibroblasts [47, 48] . Classical pharmacotherapeutic agents used to treat heart failure, including ACEi, ARBs, and aldosterone receptor blockers, increase ACE2 activity and/or expression, indicating its importance in the cardiac diseases establishment and progression [49] [50] [51] . Additionally, pharmacological and genetic (transgenic animals and gene transfer) approaches have evidenced the significance of ACE2 in cardiac pathologies. Despite some controversies concerning the consequences of the ACE2 deficiency, in general, evidences indicate a protective role of ACE2 in the heart [48, [52] [53] [54] [55] [56] [57] . Crackower and colleagues [52] were the first to demonstrate that genetic ablation of ACE2 results in severe blood-pressure-independent systolic impairment. Also, disruption of ACE2 was able to accelerate cardiac hypertrophy and shortened the transition period to heart failure in response to pressure overload by increasing local Ang II [54] . Recently, it has been demonstrated that loss of ACE2 enhances the susceptibility to myocardial infarction, with increased mortality, infarct expansion and adverse ventricular remodeling [56] . In keeping with these genetic findings, pharmacological inhibition of ACE2 exacerbated cardiac hypertrophy and fibrosis in Ren-2 hypertensive rats [58] . On the other hand, cardiac overexpression of ACE2 prevented hypertension-induced cardiac hypertrophy and fibrosis in spontaneously hypertensive rats (SHR) and in Ang-II-infused rats [59, 60] . Indeed, transfection of Lenti-ACE2 (lentivirus containing ACE2 cDNA) or Ad-ACE2 (recombinant adenovirus carrying the murine ACE2) into the surrounding area of the infarcted myocardium was protective against pathological remodeling and cardiac systolic dysfunction in a rat model of myocardial infarction [61, 62] . This effect was associated with decreased expression of ACE and Ang II and increased expression of Ang-(1-7) [62] . Collectively, these observations reveal that ACE2 effectively plays a protective role in the cardiac structure and function. Since the discovery of Ang-(1-7) in the late 1980s [63, 64] , several studies have demonstrated important effects of this peptide in hearts. The presence of Ang-(1-7) and its receptor Mas in the heart [65, 66] and the ability of this organ to produce Ang-(1-7) [55, 67] are evidences of the role of this peptide in cardiac tissues. Functionally, Ang-(1-7) induces an antiarrhythmogenic effect against ischemia/reperfusion injuries in rats [17, 68] as well as prevents atrial tachycardia and fibrillation in rats and dogs [69, 70] . Treatment with Ang-(1-7) improved the coronary perfusion and cardiac function in rats after myocardial infarction [71] and after ischemia/reperfusion injury [72] . Increases in circulating Ang-(1-7) levels in transgenic rats reduced the cardiac hypertrophy [17] and fibrosis [20, 22] induced by isoproterenol administration. These effects are apparently independent of changes in blood pressure since Grobe and colleagues [18] have demonstrated that the antifibrotic and antihypertrophic actions of Ang-(1-7) are still observed in Ang-II-infused hypertensive rats. Local overexpression of Ang-(1-7) in hearts of mice and rats improved the myocardial contractility and prevented the isoproterenol-and hypertension-induced cardiac remodeling [19, 21] . Altogether, these findings support a direct effect of Ang-(1-7) in the heart. Further evidence for the role of Ang-(1-7)/Mas in the pathophysiology of the heart came from experimental protocols utilizing mice with genetic deficiency of Mas. They revealed that the cardiac function is impaired in Mas knockout mice likely due to the increased extracellular matrix proteins deposition in the heart [66, 73] . This profibrotic phenotype may be related to changes in matrix metalloproteinases (MMPs) and tissue inhibitors of metalloproteinases (TIMPs) levels and/or activities [74, 75] . Although further elucidations regarding the signaling pathways involved in Mas activation are necessary, some mechanisms have been proposed. Overexpression of Ang-(1-7) in hearts of rats causes an improvement in the [Ca 2+ ] handling in cardiomyocytes and increases the expression of SERCA2a [21] . In keeping with these results, cardiomyocytes from Mas-deficient mice present slower [Ca 2+ ] i transients accompanied by a lower Ca 2+ ATPase expression in the sarcoplasmic reticulum [66, 76] . Although acute Ang-(1-7) treatment failed to alter Ca 2+ handling in ventricular myocytes of rats [76] , these findings suggest an important role of the Ang-(1-7)/Mas in the long-term maintenance of the Ca 2+ homeostasis in the heart. One of the mechanisms by which Ang-(1-7) plays its effects in the heart is stimulating the nitric oxide (NO) production. Indeed, it has been demonstrated that Ang-(1-7) via Mas increases the synthesis of NO through a mechanism involving the activation of the endothelial NO synthase (eNOS). These effects were abolished by A-779 and are absent in cardiomyocytes from Mas-deficient mice [76] . Recently, Gomes et al. [77] found that the treatment of isolated cardiomyocytes of rats with Ang-(1-7) efficiently prevents the Ang-II-induced hypertrophy by modulating the calcineurin/NFAT signaling cascade. These effects were blocked by NO synthase inhibition and by guanylyl cyclase inhibitors, indicating that these effects are mediated by the NO/cGMP pathway. Also, Ang-(1-7) inhibits serum-stimulated mitogen-activated protein kinase (MAPK) activation in cardiac myocytes [78] and prevents the Ang-II-mediated phosphorylation of ERK1/2 and Rho kinase in hearts in a dosedependent manner [79] . In line with these data, activation of endogenous ACE2 significantly reduced the phosphorylation of ERK1/2 in hearts of hypertensive rats (SHRs) [48] . However, Mercure et al. [19] reported that overexpression of Ang-(1-7) in hearts of rats decreases the Ang-II-induced phosphorylation of c-Src and p38 kinase, whereas the increase in ERK1/2 phosphorylation was unaffected by the expression of the transgene, thereby suggesting a selective effect of Ang-(1-7) on intracellular signaling pathways related to cardiac remodeling. Overall, these data reveal a key role of the ACE2/Ang-(1-7)/Mas axis in the pathophysiology of the cardiac structure and function. Activation of this axis might be an important strategy to develop a new generation of cardiovascular therapeutic agents against cardiac dysfunction and pathological remodeling of the heart. Early studies have reported the endothelium as the major site for generation [67] and metabolism [41] of Ang- (1) (2) (3) (4) (5) (6) (7) . In addition to Ang-(1-7), endothelial cells also express ACE2 and Mas [80, 81] . Thus, now it is recognized that the ACE2/Ang-(1-7)/Mas axis is present in vascular endothelial cells and modulates its function promoting vasorelaxation [82] , reduction of the oxidative stress [83, 84] , and antiproliferative effects [85, 86] . The vasodilatory actions of Ang-(1-7) have been reported in many studies in several vascular beds and preparations, including mouse [16, 23] and rat [15] aortic rings, canine [87] and porcine [88] coronary arteries, canine middle cerebral artery [89] , porcine piglet pial arterioles [90] , feline mesenteric vascular bed [91] , rabbit renal afferent arterioles [92] , and mesenteric microvessels of normotensive [93] and hypertensive [94] rats. Vascular Ang-(1-7) actions are still controversial in human. For example, it has been shown that Ang-(1-7) causes vasodilation in forearm circulation of normotensive subjects and patients with essential hypertension [95] while other studies were unable to report any significant effect of Ang-(1-7) in the same vascular territory in ACEi-treated patients [43] . The Mas receptor is critically involved in the vascular effects of Ang-(1-7). In fact, many of these actions are completely abolished by A-779 or partially blocked by this antagonist [3, 86, 96] . Importantly, the endothelium-dependent relaxation induced by Ang-(1-7) in mouse aortic rings is absent in vessels derived from Mas-knockout mice [16] . However, other studies have shown that Ang-(1-7) also interacts with ACE, AT 1 , and AT 2 -like receptors, suggesting the existence of additional sites of interaction for Ang-(1-7) [3, 97, 98] . Indeed, Silva et al. [99] reported evidence for the presence of a distinct subtype of Ang-(1-7) receptor sensible to D-pro 7 -Ang-(1-7), a second Mas antagonist, but not to A-779 in aortas of Sprague-Dawley rats. The vascular effects of Ang-(1-7) are endothelium dependent and involve the production of vasodilator products, such as prostanoids, NO, and endothelium-derived hyperpolarizing factor (EDHF) [16, 81, 100] . Pinheiro and coworkers [101] found that Ang-(1-7) promotes an increase in NO release in Mas-transfected chinese hamster ovary (CHO) cells [101] . Furthermore, short-term infusion of Ang-(1-7) improved the endothelial function by a mechanism involving NO release in rats [102] . Mas deletion resulted in endothelial dysfunction associated with an unbalance between NO and oxidative stress [83] . Also, Mas activation by Ang-(1-7) in human endothelial cells stimulated eNOS phosphorylation/activation via the Aktdependent pathway [81] . Other mechanisms appear to be involved in the Ang-(1-7) vascular actions. Roks et al. [103] have shown that Ang-(1-7) inhibits the vasoconstriction induced by Ang II in human internal mammary arteries, thereby suggesting that Ang-(1-7) can regulate the Ang II effects [103] . In fact, Ang-(1-7) negatively modulates the Ang II type 1 receptor-mediated activation of c-Src, and its downstream targets ERK1/2 and NAD(P)H oxidase [104] . The counterregulatory action of Ang-(1-7) on Ang II signaling has been also observed in cardiomyocytes [77] , vascular smooth muscle cells [105] , and fibroblasts [106] . Additionally, an interaction between Mas and bradykinin (Bk) type 2 (B 2 ) receptors may modulate some of the Ang-(1-7) effects in blood vessels [107] . Indeed, it has been demonstrated that Ang-(1-7) potentiates the vasodilator and hypotensive effects of Bk in several vascular beds [93, [108] [109] [110] . As the major enzyme involved in Ang-(1-7) formation, ACE2 has also a crucial role in vessels. Lovren et al. [111] have demonstrated that ACE2 ameliorates the endothelial homeostasis via a mechanism involving reduction of the reactive oxygen species production [111] . Of note, this effect was attenuated by A-779 [111] . Moreover, overexpression of ACE2 in vessels of hypertensive rats resulted in reduction in the arterial blood pressure and improvement of the endothelial function associated with increased circulating Ang-(1-7) levels [112] . Overall, these data indicate that the beneficial effects of ACE2 are, at least in part, mediated by Ang- (1-7) . Recently, we have demonstrated that activation of endogenous ACE2 causes a dose-dependent hypotensive effect in normotensive and hypertensive rats [113] . Also, the response to Bk administration was augmented in rats chronically treated with XNT, an ACE2 activator [113] . However, we were unable to demonstrate any significant effect of XNT on blood pressure in response to the administration of Ang II or Losartan in normotensive and hypertensive rats ( Figure 2 ). In the past few years, the participation of the ACE2/Ang-(1-7)/Mas axis in the establishment and progression of pulmonary diseases has become evident. Indeed, the important role of the RAS in the lung pathophysiology and the side effects and pulmonary toxicity induced by the ACEi raised the interest to evaluate the activation of the ACE2/Ang-(1-7)/Mas axis as an alternative target to treat pulmonary pathologies. Thus, it has been reported beneficial outcomes induced by the activation of this axis in animal models of acute respiratory distress syndrome (ARDS), pulmonary hypertension (PH), fibrosis, and lung cancer [31, 37, [114] [115] [116] [117] . These studies pointed out that the imbalance between the ACE/Ang II/AT 1 and the ACE2/Ang-(1-7)/Mas axes of the RAS might be relevant in lung diseases. Taking into account that systemic hypotension is an important limitation to the use of ACEi and ARBs in pulmonary patients, therapies based on the ACE2/Ang-(1-7)/Mas axis emerge as a safe and efficient approach since studies using the ACE2 activator XNT or ACE2 gene transfer have shown that these strategies induce beneficial pulmonary outcome without changes in systemic blood pressure in rats and mice [39, 117, 118] . Imai and colleagues [37] demonstrated the role of ACE2 in ARDS pathogenesis. They found that a more severe ARDS was reached in ACE2 knockout mice, and this phenotype was reversed by double genetic deletion of the ACE2 and ACE genes or by the treatment with recombinant human ACE2 (rhACE2). Furthermore, Ang II levels were related International Journal of Hypertension to the severity of the lung injury. Of note, ACE2 is widely expressed in the pulmonary endothelium, vasculature, and pneumocytes [119, 120] . Also, rhACE2 inhibited the increase of Ang II and TNF-α levels, attenuated the arterial hypoxemia and PH, and ameliorated the distribution of the pulmonary blood flow in lipopolysaccharide-induced lung injury in piglets [121] . Therefore, these studies suggest that ACE2 is a suitable target to arrest the development of ARDS in patients at risk. The stimulation of the ACE2/Ang-(1-7)/Mas axis has been successful used to prevent and reverse PH and fibrosis in animals. ACE2 activation using the compound XNT or induction of ACE2 overexpression by gene transfer efficiently prevented and, more importantly, reversed the increase of the right systolic ventricular pressure (RSVP), pulmonary fibrosis, imbalance of the RAS, and inflammation in animals (rats and mice) with PH induced by monocrotaline (MCT) or in rats with pulmonary fibrosis caused by bleomycin treatment [39, 117, 118] . In keeping with these findings, Ang-(1-7) gene transfer into the lungs triggered similar protective actions in MCT-treated rats [39] . In addition, Ang-(1-7) via Mas prevented the apoptosis of alveolar epithelial cells and the Jun N-terminal kinase (JNK) activation induced by bleomycin [122] . The involvement of the Ang-(1-7)/Mas in PH was further evidenced by the observation that the XNT effects are blocked by A-779 [117] . Furthermore, in both lung specimens from patients with idiopathic pulmonary fibrosis and from animals with bleomycin-induced pulmonary fibrosis were reported a reduction in mRNA, protein, and activity of ACE2 with a reciprocal increase in Ang II level [116] . A growing body of studies has focused on the relevance of the ACE2/Ang-(1-7)/Mas axis in the pulmonary cancer pathophysiology. The protein expression of ACE2 is reduced in non-small-cell lung carcinoma (NSCLC) along with an increase in Ang II levels. Moreover, overexpression of ACE2 in cultured A549 lung cancer cells and in human lung cancer xenografs inhibited the cell growth and the vascular endothelial growth factor-a (VEGFa) expression induced by Ang II [123, 124] . Gallagher and Tallant [125] evaluated the effects of several angiotensin peptides [Ang I, Ang II, Ang-(2-8), Ang- (3) (4) (5) (6) (7) (8) , and Ang-(3-7)] in SK-LU-1 cancer cells growth, and only Ang-(1-7) showed significant attenuation of the DNA synthesis and proliferation. The antiproliferative effect of Ang-(1-7) was mediated by its receptor Mas and inhibition of the ERK1/2 pathway. Neither the blockage of AT 1 nor AT 2 succeeded in inhibiting the action of Ang-(1-7). In keeping with these data, the antiproliferative effect of Ang-(1-7) was observed in human A549 lung tumor xenograft growth along with a marked decrease in the vessel density in mice through a mechanism involving cyclooxygenase-2 (COX-2) [126, 127] . Of note, in a nonrandomized phase I clinical trial conducted by Petty and colleagues [38] , subcutaneous injections of Ang-(1-7) were administered in 18 patients with advanced solid tumors refractory to standard therapy. Despite the mild adverse effects observed with the Ang-(1-7) treatment, generally it was well tolerated. There were no treatment-related deaths. Clinical benefits were observed in 27% of the patients. Altogether, these studies provide insights into the involvement of the ACE2/Ang-(1-7)/Mas axis in lung cancer. Many advances have been achieved regarding the therapeutic regulation of the RAS. Current therapies based on the modulation of the RAS include the ACEi, ARBs, and renin inhibitors. In general, these drugs prevent or reverse endothelial dysfunction and atherosclerosis, reduce cardiovascular mortality and morbidity of patients with coronary artery disease, and hold antihypertensive effects [128] . Classically, the mechanisms of action of the ACEi and ARBs involve the blockade of the synthesis and actions of Ang II, respectively. However, the RAS is a complex hormonal system and, consequently, other mechanisms are likely implicated in the actions of these drugs [42, 86, 129] . They cause substantial increase in plasma levels of Ang-(1-7), leading to the assumption that their clinical effects might be partly mediated by this heptapeptide [42, 130] . Indeed, a variety of effects of the ACEi and ARBs can be abolished or attenuated by Mas antagonism, confirming the role of Ang-(1-7) in the actions of these compounds [129, 131] . The beneficial effects of Ang-(1-7) as well as its likely involvement in the effects of the ACEi and ARBs represent a strong evidence for the therapeutic potential of the activation of the ACE2/Ang-(1-7)/Mas axis (Figure 3 ). Ang-(1-7) Formulations. The beneficial effects of Ang-(1-7) are well known; however, the therapeutic utilization of this peptide is limited due to its unfavorable pharmacokinetic properties. Ang-(1-7) has a short half-life (approximately 10 seconds) since it is rapidly cleaved by peptidases [132] . Furthermore, Ang-(1-7) is degraded during its passage through the gastrointestinal tract when orally administrated. Thus, new strategies are crucial to make feasible the clinical application of Ang-(1-7) . Recently, a formulation based on the Ang-(1-7) included into hydroxypropyl β-cyclodextrin [HPβCD/Ang-(1-7)] was developed by Lula and colleagues [133] . Cyclodextrins are pharmaceutical tools used for design and evaluation of drug formulations, and they enhance the drug stability and absorption across biological barriers and offer gastric protection [134] . The amphiphilic character of cyclodextrins allows the possibility of formation of supramolecular inclusion complexes stabilized by noncovalent interactions with a variety of guest molecules [133, 134] . In this regard, the formulation HPβCD/Ang-(1-7) allowed the oral administration of Ang-(1-7). Pharmacokinetic and functional studies showed that oral HPβCD/Ang-(1-7) administration significantly increases plasma Ang-(1-7) levels and promotes an antithrombotic effect that was blunted in Mas deficient mice [135] . Marques and colleagues [136] have found that chronic oral administration of HPβCD/Ang-(1-7) significantly attenuates the heart function impairment and cardiac remodeling induced by isoproterenol treatment and myocardial infarction in rats [136] . In addition, liposomal delivery systems represent an alternative method to administer Ang-(1-7) [137] . Administration of liposomes containing Ang-(1-7) in rats led to prolonged hypotensive effect for several days in contrast to the response observed when the free peptide was used [137, 138] . A strategy used to protect the Ang-(1-7) against proteolytic degradation was proposed by Kluskens and coworkers [139] . Using the ability of prokaryotes to cyclize peptides, they synthesized a cyclic Ang-(1-7) derivative [thioetherbridged Ang-(1-7)] which presented an increased stability in homogenates of different organs and plasma and enhanced the Ang-(1-7) bioavailability in rats [139] . Furthermore, cyclized Ang-(1-7) induced a relaxation in precontracted aorta rings of rats which was blocked by the Ang-(1-7) receptor antagonist D-Pro 7 -Ang-(1-7), providing evidence that cyclized Ang-(1-7) also interacts with Mas [139] . Agonists. AVE 0991 was the first nonpeptide synthetic compound developed with the intention of stimulating the Mas receptor. This compound mimics the Ang-(1-7) effects in several organs such as vessels [140, 141] , kidney [101] , and heart [142, 143] . Similar to Ang-(1-7), AVE 0991 induced a vasodilation effect which was absent in aortic rings of Mas-deficient mice [140] . Moreover, its effects in aortic rings were blocked by the two Ang-(1-7) receptor antagonists, A-779 and D-Pro 7 -Ang-(1-7) [140] . AVE 0991 potentiated the acetylcholine-induced vasodilation in conscious normotensive rats, and this effect was abolished by A-779 and L-NAME [102] . Similarly, it was able to increase the hypotensive effect of Bk in normotensive rats, and A-779 also blocked this effect [107] . Ferreira et al. [142, 143] reported that AVE 0991 protects the heart against cardiac dysfunction and remodeling caused by isoproterenol treatment or by myocardial infarction in rats [142, 143] . In Mas-transfected cells, AVE 0991 induced NO release which was blunted by A-779 and not by AT 2 or AT 1 antagonists [101] . All these data support the concept that AVE 0991 is an Ang-(1-7) mimetic and that its actions are mediated by the interaction with Mas. Using a computational discovery platform for predicting novel naturally occurring peptides that may activate GPCR, two novel peptides, designated as CGEN-856 and CGEN-857, with amino acid sequence unrelated to angiotensin peptides, were found to display high specificity for Mas [23] . These peptides elicited Ca +2 influx in CHO cells overexpressing Mas without any activity in AT 1 or AT 2 receptors [144] . CGEN-856S, a derivative of the CGEN-856 peptide, induced beneficial cardiovascular effects similar to those caused by Ang-(1-7) [23] . This compound competes with Ang-(1-7) for the same bind site in Mas-transfected cells. Furthermore, similar to Ang-(1-7), CGEN-856S produced a vasodilation effect which was absence in Mas-deficient mice, indicating that this compound also acts via Mas [23] . This was confirmed by the inhibition of the CGEN-856S effects by the Mas antagonist A-779. Importantly, Savergnini et al. [23] showed that CGEN-856S promotes antiarrhythmogenic effects and produces a small dose-dependent decrease in arterial pressure of conscious SHR [23] . A new approach addressing the therapeutic potential of the activation of the ACE2/Ang-(1-7)/Mas axis was proposed by Hernández Prada et al. [113] . Based on the crystal structure of ACE2 and using a virtual screening strategy, it was identified small molecules that may interact with this enzyme leading to changes in its conformation and, consequently, enhancing its activity [113] . Thus, the ACE2 activator, namely XNT, was identified and its administration in SHR decreased blood pressure, induced an improvement in cardiac function, and reversed the myocardial and perivascular fibrosis observed in these animals [48, 113] . The beneficial effects of XNT were also observed in rats with PH induced by MCT [117] . Furthermore, this compound attenuated the thrombus formation and reduced the platelet attachment to vessels in hypertensive rats [145] . It appears that the pharmacological activation of ACE2 promotes its beneficial effects due to an increased Ang-(1-7) production with concomitant degradation of Ang II. In fact, coadministration of A-779 abolished the protective effects of XNT on PH [117] . In addition, the antifibrotic effect of XNT observed in hearts of SHR was associated with increases in cardiac Ang-(1-7) expression [48] . However, it is also pertinent to point out that off-target effects of XNT on these beneficial outcomes cannot be ruled out at the present time. The complexity of the RAS is far beyond we could suspect few years ago. There is growing evidence that changes in the novel components of the RAS [Ang-(1-7), ACE2, and Mas] may take part of the establishment and progression of cardiovascular and respiratory diseases. Importantly, these new components of the RAS, due to their counter regulatory actions, are candidates to serve as a concept to develop new cardiovascular and respiratory drugs.